E hnic and Religious Pola iza ion and Social
Con lic
Joan Es eban
∗
Lau a Mayo al
†
Janua y 13, 2011
Fi s e sion No embe 2, 2009
Abs ac
In his pape we examine he link be ween e hnic and eligious pola iza-
ion and con lic using in e pe sonal dis ances o e hnic and eligious a i udes
ob ained om he Wo ld Values Su ey. We use he Duclos e al (2004) pola -
iza ion index. We measu e con lic by means on an index o social un es , as
well as by he s anda d con lic onse o incidence based on a h eshold numbe
o dea hs. Ou esul s show ha aking dis ances in o accoun signi ican ly
imp o es he quali y o he i . Ou measu e o pola iza ion ou pe o ms he
measu e used by Mon al o and Reynal-Que ol (2005) and he ac ionaliza ion
index. We also ob ain ha bo h e hnic and eligious pola iza ion a e signi ican
∗
Ins i u d’Anàlisi Económica, CSIC, and Ba celona GSE; email: joan.es [email protected]. Fi-
nancial suppo om he Axa Resea ch Fund, he Go e nmen o Ca alonia, and he CICYT p ojec
n. SEJ2006-00369 is g a e ully acknowledged.
†
Ins i u d’Anàlisi Económica, CSIC; email: lau a.mayo [email protected]. Membe o he Ba celona
GSE Resea ch Ne wo k unded by he Go e nmen o Ca alonia. Financial suppo om he CICYT
p ojec n. SEJ2006-00369 is g a e ully acknowledged.
1
in explaining con lic . The esul s imp o e when we use an indica o o social
un es as he dependen a iable.
JEL-Classi ica ion:
Key-wo ds: con lic , pola iza ion, ac ionaliza ion, e hnici y, eligion. Acknowledgemen s:
Ignacio O uño-O ín has g ace ully compu ed o us he linguis ic dis ances be ween g oups
based on he e hnologue da ase . We a e g a e ul o Bo ek Vasicek o a highly compe en
assis ance.
2
1. INTRODUCTION
In his pape we examine he link be ween e hnic and eligious pola iza ion and
con lic . Income inequali y has adi ionally been seen as a majo po en ial cause o
con lic . Ea ly empi ical s udies ocussed on he pe sonal dis ibu ion o income o
o landowne ship.
1
Howe e , as he su ey a icle by Lichbach (1989) concluded, he
empi ical esul s ob ained we e lacking o signi icance and ambiguous. This appa en
lack o connec ion be ween economic inequali y and social con lic has been possibly
due o he ac ha mos o he domes ic con lic s since 1945 ha e had a s ong
e hnic/ eligious componen . Indeed, Ho owi z (1985) al eady no ed ha he Ma xian
p ophecy o an ine i able class s uggle has ended up ha ing an e hnic ul illmen .
2
Following he con ibu ion by Eas e ly and Le ine (1997) he a en ion o empi ical
esea ch has shi ed owa ds he e hnic and eligious social di ides as a cause o con lic
and low collec i e ac ion. The index o ac ionaliza ion,F, has been he mos widely
used measu e o he e hnic/ eligious composi ion o a coun y. Bu o he indica o s
such as he Gini-G eenbe g
3
index G o he pola iza ion indices by Es eban and Ray
(1994) Pand by Reynal-Que ol (2002) RQ
4
ha e been used as well. The empi ical
wo ks by Alesina e al. (1999), Alesina e al. (2003), Alesina and La Fe a a (2005),
Collie (2001), Collie and Ho le (2004), Desme e al (2009,2010), Fea on (2003),
Fea on and Lai in (2003), Miguel e al. (2004), Mon al o and Reynal-Que ol (2005)
1
See he wo ks o B ocke (1992), Midla ski (1988), Mulle and Seligson (1987), Mulle e al.
(1989), and Nagel (1974), among o he s.
2
See Es eban and Ray (2008) on he salience o e hnici y o e class in social con lic .
3
See Desme e al. (2009,2010) on G eenbe g’s index.
4
See Es eban and Ray (1994) –also Wol son (1994)– o he ea lies pola iza ion measu e, ER,
and Reynal-Que ol (2002) o a special case o he Es eban-Ray measu e, RQ. Duclos e al. (2004)
ex end Es eban and Ray’s measu e o con inuous dis ibu ions. See also he special issue o he
Jou nal o Peace Resea ch edi ed by Es eban and Schneide (2008) de o ed o he links be ween
pola iza ion and con lic .
1
and Reynal-Que ol (2002) a e ep esen a i e o his li e a u e.
Mos o he empi ical wo k has been based on he indices o ac ionaliza ion Fand
pola iza ion RQ. Bo h sha e he ea u e ha a e based on g oup sizes only and do
no make use o a ia ions in in e -g oup dis ances. Fea on (2003) al eady made he
poin ha e hno-linguis ic dis ances do play a key ole in explaining e hnic con lic and
compu ed a measu e based on dissimila i y be ween pai s o languages. Mon al o and
Reynal-Que ol (2005) (MRQ he eina e ) dismissed he use o such dis ances a guing
ha , using Fea on’s da a, he co ela ion be ween G( ha uses dis ances) and F( ha
doesn’ ) is 0.82. Howe e , his claim has been ecen ly challenged by Desme e al
(2009,2010). They e-examine he poin made in a ious pape s by Alesina conce ning
he lowe le el o social ans e s in e hnically he e ogeneous socie ies and ind ha
he measu es ha include dis ances ou pe o m he ones ha don’ . Speci ically,
hey ob ain ha Gis highly signi ican while Fisn’ and ha he same is ue o P
ela i e o RQ.
5
Ou pape con ibu es o he li e a u e on wo coun s. Fi s , ou pape s udies
he link be ween con lic and e hnic and eligious pola iza ion using in e pe sonal
dis ances d i en by he in ensi y o he e hnic and eligious a i udes ob ained om
he Wo ld Values Su ey. We de i e he in ensi y o eelings by agg ega ing he
answe s o a se o ques ions ela ed o eligious o e hnic a i udes. This pe mi s us
o compu e a pola iza ion measu e ha depend on in e -pe sonal dis ances, such as
P, and es whe he i pe o ms be e han he ones ha do no use his in o ma ion,
such as Fand RQ.
Second, oge he wi h he s anda d p ac ice o dicho omizing he occu ence o con-
5
A second ele an ea u e o his li e a u e is ha i has been gea ed owa ds inding empi ical
egula i ies a he han es ing he implica ions o a speci ic model o con lic . In Es eban, Mayo al
and Ray (2010) we es he empi ical implica ions o he con lic model se up in Es eban and Ray
(2010), as a i s s ep owa ds an explici link be ween heo y and ac s.
2
lic –wa /peace– depending on he numbe o dea hs exceeding a gi en h eshold,
we also use a second, con inuous indica o o social un es based on poli ical assassi-
na ions, demons a ions, s ikes, poli ical p isone s, e c. This pe mi s us o o e come
wo majo empi ical p oblems in he li e a u e ha we will discuss in Sec ion 4: (i)
he esul s may depend on he choice o he h eshold le el; and (ii) he dilemma
be ween onse e sus incidence as de ini ions o con lic .
Ou empi ical exe cise di ec ly compa es he pe o mance o e hnic and eligious
pola iza ion as measu ed by RQ and P, using he same con ols as in MRQ. In all
ou es ima ions we use he con inuous index o in ensi y o con lic as well as he
classic bina y measu e based on a h eshold le el o casual ies. We also check o
po en ial endogenei y in case he in ensi y o a i udes is he consequence a he
han he cause o con lic . We inally pe o m a se ies o obus ness es s. Ou
esul s s ongly suppo he hypo hesis ha in ensi y o eelings is highly signi ican ,
mos especially o e hnic a i udes. Mo e impo an ly, he dis ance sensi i e indices
o eligious and e hnic pola iza ion, P, a e bo h independen ly signi ican . When
we simul aneously conside he wo o hem, bo h a e signi ican when we use he
disc e e measu e o con lic , bu eligious pola iza ion ceases o be signi ican wi h
he con inuous indica o o con lic . In all cases, he use o in ensi y o a i udes
signi ican ly inc eases he explana o y powe o he model, as we ob ain le els o R
2
ha a e much highe han usual in his li e a u e.
The pape is o ganized as ollows. In he nex sec ion we summa ize he main
ea u es o he a ious dis ibu ional indices ha ha e been used in he li e a u e.
Sec ion discusses ou app oach o he measu emen o in e -pe sonal dis ances, key
o ou exe cise. Sec ion desc ibes in de ail he da a used in bo h he main exe cise
and he di e en obus ness es s pe o med. Sec ion p esen s ou empi ical esul s
and Sec ion concludes.
3

2. DISTRIBUTIONAL INDICES
The e a e a a ie y o indica o s cap u ing di e en ea u es o a dis ibu ion. We
ha e al eady men ioned ou di e en indices ha ha e been used in empi ical wo k:
F,G,RQ and P.
6
Wha a e he app op ia e indica o s o be used i we wan o
p edic con lic ?
This appa en he e ogenei y o measu es can be p esen ed as di e en ways o
speci ying he measu emen o in e pe sonal dis ances and he weigh o g oup size.
On he i s dimension, he speci ica ion o dis ances de ines wo classes o measu es.
One class e ains he measu ed in e -pe sonal dis ances while a second class conside s
all o he g oups equally dis an – his common dis ance is no malized o uni y. G
and Pbelong o he o me class and Fand RQ o he la e . The second dimension
e e s o he ea men o g oup sizes. We ha e he e oo wo classes o measu es.
One class does no ake in o accoun he e ec o g oup size on he sense o iden i y.
Fand Gbelong o his class. The second ype assigns “ e u ns” o g oup size and
ans o ms he own g oup size o a powe . This is he ea u e ha makes pola iza ion
measu es Pand RQ dis inc ly di e en om inequali y measu es.
The iden i y/aliena ion app oach o social an agonisms in oduced by Es eban and
Ray (1994) may help o es ablish a axonomy o e he a ie y o dis ibu ional indices
ha ha e been used so a . Acco dingly wi h his app oach, in e pe sonal an agonism
is he conjoin esul o he sense o iden i ica ion wi h one’s own g oup and he
aliena ion el owa ds membe s o o he g oups. The sense o iden i y depends on
he size o one’s g oup, n
i
, and he eeling o aliena ion on he pe cei ed dis ance
be ween g oups iand j,d
ij
.
Mo e o mally, Es eban and Ray (1994) and Duclos, Es eban and Ray (2004) s a
6
Collie and Hoe le (2004) ha e also used he a io o he la ges o e he second g oup and
Desme e al (2009,2010) in oduce he index o pe iphe al he e ogenei y.
4
by de ining he gene al class o measu es o socie al an agonism, A. The an agonism
el by a membe o g oup i is-a- is a membe o g oup j,a(i, j)can be exp essed as
a(i, j) = φ(n
i
, d
ij
),(1)
wi h φ(n
i
, d
ii
) = 0. To al socie al an agonism is de ined as he sum o all in e pe sonal
an agonisms:
A=
i

j
n
i
n
j
φ(n
i
, d
ij
).(2)
Es eban and Ray (1994) embody he concep o pola iza ion o a dis ibu ion in a se
o i e axioms. F om hese axioms, hey uniquely de i e he measu e o pola iza ion
P=
i

j
n
1+α
i
n
j
d
ij
,(3)
wi h 1≤α1.6.
7
No e ha Pis a speci ic measu e o socie al an agonism. The a o emen ioned
axioms imply ha in e pe sonal an agonism is o he o m:
φ(n
i
, d
ij
) = n
α
i
d
ij
.
Hence, pola iza ion cap u es bo h componen s o in e pe sonal an agonism: iden-
i y and aliena ion.
Suppose now ha we simply posi ha in e pe sonal an agonism does no depend
on he sense o iden i y. Then φ(n
i
, d
ij
) = d
ij
. We ob ain he Gini-G eenbe g index
G=
i

j
n
i
n
j
d
ij
.(4)
7
Es eban and Ray (2010) in oduce an axiom ha combined wi h he axioms o con inuous
dis ibu ions in Duclos e al (2004) pins down α= 1.
5
I in addi ion we posi ha a shi in aliena ion, ˜
d
ij
=d
ij
+δ o all i=j, does no
modi y in e pe sonal an agonism, hen φ(n
i
, d
ij
) = k o all i=j. No malizing k= 1
we ob ain he Hi schman-He indahl ac ionaliza ion index
G=
i

j=i
n
i
n
j
=
i
n
i
(1 −n
i
).(5)
Finally, i we make he p e ious assump ion conce ning aliena ion, bu con inue
o e ain he ole o iden i ica ion, we ob ain ha φ(n
i
, d
ij
) = n
α
i
. This gi es us he
measu e p oposed by Reynal-Que ol (2002):
RQ =
i

j=i
n
1+α
i
n
j
=
i
n
1+α
i
(1 −n
i
).
8
(6)
The e o e, he esul by MRQ ha RQ is signi ican in explaining con lic while
Fis no , can be in e p e ed as indica ing ha g oup conce n –hence he g oup-size
e ec – is impo an . Ou pape can be seen as a es ha no only g oup size bu also
aliena ion bo h ma e o social con lic . To his e ec , we shall show ha Phas a
signi ican ly highe explana o y powe o con lic han any o he o he dis ibu ional
measu es. Because in e pe sonal dis ances d
ij
do play a signi ican complemen a y
ole we shall show ha Pou pe o ms RQ as an independen explana o y a iable
o con lic .
8
No e ha his measu e is concep ually closely ela ed o F. Indeed, while F ells us he p oba-
bili y ha wo pe sons d awn andomly belong o di e en g oup, RQ wi h α= 1 – he empi ically
ele an a ian – ells us he p obabili y ha ou o h ee people wo belong o he same g oup and
he hi d o any o he g oup.
6
3. IDENTITY AND ALIENATION IN ETHNIC AND RELIGIOUS
POLARIZATION
One o he dis inc ea u es o ou exe cise is he use o pola iza ion indices ha
depend on bo h, g oup size and in e -pe sonal dis ance. As men ioned be o e, he
p e ious wo k by MRQ, while emphasizing he ole o g oup sizes, dis ega ded in e -
pe sonal dis ances. Indeed, he measu e RQ only depends on he popula ion size o
he di e en e hnic/ eligious g oups.
The classi ica ion o he popula ion in o e hnic and/o eligious g oups is no as
s aigh o wa d as i i s appea s. I poses he p oblem o he de ini ion o he g oups.
E en when he asc ip ion o indi iduals o g oups is unequi ocal, he e emains he
issue o wha a e he ele an g oups. To illus a e he poin , le us ake he line
iden i ying e hnic g oup wi h language. E hnologue eco ds 6,912 di e en languages
wo ldwide. This gi es an a e age o hi y i e e hnolinguis ic g oups pe coun y [ o
he 195 coun ies exis ing oday]. In India E hnologue iden i ies 415 languages.
9
I
is plain ha one needs o “agg ega e" o e hese mic o-g oups and ocus on b oade
de ini ions, me ging a ious “simila " e hnic g oups. The same a gumen can be made
o eligions.
10
And his akes us o a main p oblem we wish o unde sco e. This is he issue o
he in e -pe sonal o in e -g oup dis ances. The way he g ouping is pe o med in
he li e a u e implici ly assumes ha in e -g oup dis ances a e ei he ze o –when
he g oups a e me ged in o one– o uni y –when hey a e conside ed alien o each
o he . This p oblem can be bypassed by using in e -g oup dis ances. The dis ances
among e hnolinguis ic g oups ha e been compu ed by Fea on (2003) and by Desme
e al. (2009,2010) on he basis o di e en measu es o linguis ic simila i y. The use o
hese in e -g oup measu es pe mi s o go beyond he dycho omic [0,1] dis ance and
9
Al hough he 1991 census ecognizes 1,576 “mo he ongues".
10
Fo ins ance,MRQ ea “ch is ians" as a single g oup.
7
4.3. Addi ional Independen Va iables
While he dependen a iable is he incidence o con lic in each pe iod, he con ol
a iables e e o he i s yea o each pe iod o a e by hei na u e in a ian in ime.
The numbe o coun ies a ies sligh ly wi h he exe cise and is indica ed oge he
wi h each empi ical esul .
The con ol a iables a e as ollows. Sociopoli ical a iables: size o he popula ion,
le el o democ acy. Economic a iables: eal GDP pe capi a, LGDP C, sha e o
p ima y expo s on GDP, PRIMEXP , dependence on oil expo s, OIL. Geog aphic
a iables: pe cen age o moun ainous e ain, MOUNT , noncon inguency o coun y
e i o y, NONCONT and egional dummies o La in Ame ica, D−AMER, Asia,
D−ASIA, and Sub-Saha an A ica, D−AFRICA. While he jus i ica ion o each
con ol a iable can be ound elsewhe e (Fea on and Lai in, 2003, Collie and Hoe le ,
2004, 2008, Miguel e al., 2004, MRQ) he de ails a e p o ided in Table B1 included
in Appendix B. The da a on eligious and e hnic composi ion o coun ies used o
calcula ion o Fand RQ measu es in he main ex ha e been aken om MRQ.
4.4. Sample Size
The a ailabili y o da a om di e en sou ces se e ely condi ions he size o ou
sample. Le us s a by he WVS. I has been conduc ed only o limi ed numbe
o coun ies. The sample o coun ies has expanded om 20 coun ies pa icipa ing
in he i s wa e in 1981 o a o al o 97 coun ies being su eyed in he la es i h
wa e. Mo eo e , ou analysis is es ic ed o hose coun ies whe e a sui able se o
ques ions ela ed o eligious o e hnic ole ance we e asked and whe e eligious o
e hnic iden i ica ion o each esponden was a ailable .
A main poin o ou pape is ha he dis ance-based Ppola iza ion index ou pe -
o ms he RQ index wi h no a ia ion in in e -pe sonal dis ances. The WVS sample
14

does no coincide wi h he sample used by MRQ. The di ec compa abili y o ou
esul s wi h hose ob ained by MRQ limi s e en u he he numbe o coun ies in
ou sample. While hei sample includes 117 coun ies o he pe iod 1960-1999, he
o e lap wi h he WVS yields 51 and 61 coun ies o e hnic and eligious da a, e-
spec i ely. When combined wi h he numbe o i e-yea pe iods he sample con ains
385 and 464 obse a ions. espec i ely. This subse o coun ies p esen s ea u es o
con lic simila o he la ge se . Using he PRIOCW a iable, Mon al o and Reynal-
Que ol’s (2005) da ase ea u es 159 pe iods wi h ongoing ci il wa , while we ha e 83
pe iods o he eligious subse o coun ies and 75 o he e hnic one. I we compu e
he mean incidence o wa pe pe iod ( he numbe o wa pe iods di ided by he o al
sample) we ob ain e y simila esul s. Fo he e hnic sample we ha e a mean inci-
dence o 0.145 and 0.157 o he eligious sample, while Mon al o and Reynal-Que ol
eco d a mean incidence o 0.168.
When we use ISC as he dependen a iable we es ic o he same se o coun ies.
Un o una ely, he in o ma ion on he ISC index is missing o some pe iods. This
educes he o al numbe o obse a ions o 344 and o 406 o he e hnic and eligious
da a.
When pe o ming he obus ness checks we shall be using di e en and la ge sam-
ples, including mo e coun ies and mo e pe iods. The speci ics a e desc ibed in de ail
when needed.
5.EMPIRICAL RESULTS: ETHNIC AND RELIGIOUS
POLARIZATION AND CONFLICT
5.1. Pola iza ion and Con lic
We use he ob ained indi idual in ensi y o eelings o compu e he Pindex o
pola iza ion. We wish o es whe he Pou pe o ms RQ in explaining con lic . We
15
conside he same se o independen a iables as in MRQ and wo dependen ones:
he bina y a iable o incidence o ci il wa , PRIOCW, and he con inuous measu e
ha cap u es he in ensi y o social con lic , ISC. Resul s a e epo ed in Tables 1 and
2, espec i ely. Columns 1-3 e e o e hnic pola iza ion, columns 4-6 e e o eligious
pola iza ion. Fo each case, he i s column ep oduces he esul s when only he RQ
index is included in he educed sample (columns (1) and (4)), he second column uses
Pins ead and he hi d column uses bo h, RQ and P, pe mi ing a di ec con as
be ween he measu es wi hou and wi h in ensi y o a i udes. Columns (7) and (8)
in oduce bo h eligious and e hnic P and he ou indices conside ed in hese exe cise,
espec i ely.
16
T 1. P  C: P RQ
Va iable (1) (2) (3) (4) (5) (6) (7) (8)
LGDPC −0.457
(0.084)
−0.564
(0.050)
−0.563
(0.048)
−0.290
(0.374)
−0.162
(0.523)
−0.143
(0.616)
−0.450
(0.162)
−0.576
(0.083)
LPOP 0.124
(0.576)
0.095
(0.650)
0.095
(0.658)
0.164
(0.448)
0.160
(0.457)
0.153
(0.506)
0.045
(0.846)
0.073
(0.738)
PRIMEXP −0.632
(0.776)
−1.129
(0.595)
−1.116
(0.601)
−0.600
(0.796)
−1.705
(0.564)
−1.766
(0.569)
−2.848
(0.301)
−2.443
(0.373)
MOUNT 0.006
(0.700)
0.015
(0.290)
0.015
(0.363)
0.013
(0.225)
0.020
(0.110)
0.020
(0.099)
0.032
(0.021)
0.033
(0.061)
NONCONT 0.282
(0.696)
0.613
(0.434)
0.609
(0.452)
0.176
(0.784)
0.449
(0.489)
0.457
(0.475)
0.159
(0.154)
1.178
(0.134)
DEM −0.148
(0.768)
−0.157
(0.723)
−0.157
(0.723)
0.027
(0.954)
0.160
(0.724)
0.157
(0.725)
−0.319
(0.463)
−0.283
(0.548)
RQ
E h
0.737
(0.499)
-−0.040
(0.975)
- - - - −0.028
(0.983)
RQ
Rel
- - - 1.071
(0.246)
-0.164
(0.860)
-−1.108
(0.311)
P
E h
-2.849
(0.013)
2.853
(0.014)
- - - 3.920
(0.002)
4.292
(0.002)
P
Rel
- - - - 2.343
(0.009)
2.2858
(0.015)
2.059
(0.038)
2.550
(0.023)
Pseudo-R
2
0.066 0.177 0.177 0.084 0.139 0.139 0.253 0.257
Coun ies 51 51 51 61 61 61 50 50
Obse a ions 385 385 385 464 464 464 377 377
No es
:
The dependen a iable is PRIOCW and he es ima ion me hod is logi . P- alues a e epo ed
in b acke s. Robus s anda d e o s adjus ed o clus e ing ha e been employed o compu e z-s a is ics.
Table 1 shows ha while RQ is no signi ican in ou educed sample o coun ies
(columns (1) and (4)), bo h e hnic and eligious Pa e so (columns (2) and (5)).
When bo h P and RQ a e included in he eg ession (columns (3) and (6)), he
signi icance o P emains and he es ima ed coe icien s a e e y simila as when only
Pis conside ed. These conclusions do no change i bo h e hnic and eligious P
a e included in he eg ession (column 7) o when he ou indices conside ed in his
17
exe cise a e simul aneously in oduced (column 8). Fu he mo e, he in oduc ion
o Pinc eases he pseudo-R
2
subs an ially. When bo h eligious and e hnic P a e
conside ed, he pseudo-R
2
eaches 0.25.
We now eplica e he same exe cise epo ed in Table 1 on he pe o mance o P
ela i e o RQ using as dependen a iable he con inuous indica o o con lic , ISC.
The esul s a e displayed in Table 2. This able is o ganized exac ly as Table 1.
T 2. P  C: P  RQ  ISC
Va iable (1) (2) (3) (4) (5) (6) (7) (8)
LGDPC −0.483
(0.000)
−0.502
(0.000)
−0.498
(0.000)
−0.501
(0.000)
−0.431
(0.000)
−0.446
(0.000)
−0.481
(0.000)
−0.567
(0.000)
LPOP 0.139
(0.078)
0.118
(0.120)
0.129
(0.092)
0.191
(0.016)
0.174
(0.019)
0.178
(0.016)
0.109
(0.154)
0.122
(0.104)
PRIMEXP −1.144
(0.112)
−1.154
(0.097)
−1.169
(0.096)
−0.986
(0.202)
−1.286
(0.091)
−1.257
(0.104)
−1.341
(0.053)
−1.274
(0.062)
MOUNT 0.001
(0.874)
0.006
(0.285)
0.004
(0.553)
0.003
(0.521)
0.005
(0.305)
0.005
(0.310)
0.008
(0.182)
0.006
(0.342)
NONCONT 0.679
(0.034)
0.698
(0.023)
0.751
(0.021)
0.553
(0.028)
0.637
(0.011)
0.634
(0.011)
0.747
(0.015)
0.829
(0.010)
DEM 0.079
(0.694)
0.118
(0.524)
0.106
(0.558)
0.054
(0.793)
0.086
(0.646)
0.092
(0.623)
0.150
(0.441)
0.179
(0.372)
RQ
E h
0.680
(0.079)
-0.475
(0.198)
- - - - 0.534
(0.153)
RQ
Rel
- - - 0.186
(0.631)
-−0.121
(0.735)
-−0.559
(0.118)
P
E h
-0.921
(0.000)
0.867
(0.000)
- - - 0.878
(0.002)
0.903
(0.002)
P
Rel
- - - - 0.793
(0.026)
0.843
(0.032)
0.337
(0.189)
0.512
(0.075)
Pseudo-R
2
0.237 0.286 0.291 0.268 0.281 0.282 0.289 0.304
Coun ies 51 51 51 61 61 61 50 50
Obse a ions 344 344 344 406 406 406 336 336
No es
:
The dependen a iable is ISC and he es ima ion me hod is pooled OLS. P- alues a e epo ed
in b acke s. Robus s anda d e o s adjus ed o clus e ing ha e been employed o compu e z-s a is ics.
The quali a i e conclusions o his able a e in line wi h he ones ob ained wi h
18
he disc e e a iable o con lic . Resul s in columns (1) and (4) a e e y simila as
hose in he o iginal pape by MRQ: he e hnic dimension RQ is signi ican (a he 10
pe cen le el) while he eligious one is no . E hnic P u ns ou o be signi ican in all
he speci ica ions conside ed and, when bo h e hnic P and RQ a e in oduced, e hnic
RQ ceases o be so (column (3)). On he eligious dimension we ob ain ha while RQ
is ne e signi ican , eligious P is so when in oduced alone o oge he wi h eligious
RQ (columns (5) and (6), espec i ely). Howe e , conside ing bo h he e hnic and he
eligious P indices in he eg ession limi s he signi icance o he la e (columns (7)
and (8)). Finally, he explana o y powe o he model inc eases subs an ially when
he P indices a e conside ed, as shown by he pseudo R
2
ha eaches 0.29.
5.2. Endogenei y o A i udes: IV eg essions
We ha e al eady men ioned ha ou p e ious exe cise could be objec ed on he ba-
sis o endogenei y. Di e si y measu es ha do no inco po a e in e pe sonal dis ances,
such as RQ and F, ha e usually been conside ed as exogeneous in c oss-coun y e-
g essions, he eason being ha g oup sha es a e hough o be e y s able o e ime
and small changes only ha e a mino impac on hese measu es. E hnic RQ o F could
o some ex en be endogeneous (con lic can al e g oup sha es o he de ini ions o
e hnic g oups can change h ough ime as a unc ion o economic-poli ical a iables)
bu , as poin ed ou by Alesina e al. (2003), e hnic composi ions display emendous
ime pe sis ence and hus, he exogenei y assump ion could be easonable a he 20-
30 yea ho izon ha cha ac e izes c oss coun y eg essions in his a ea. Religious
indices, howe e , may be mo e p oblema ic. In some ep essi e egimes, non-o icial
eligions migh be p osecu ed, making i di icul o membe s o hese eligions o
be coun ed as such. This may c ea e a spu ious co ela ion be ween lack o poli ical
eedom and eligious di e si y ha could bias he es ima es.
Ne e heless, he e is li le doub ha in oducing people’s a i udes in ou po-
19

la iza ion measu es makes he endogenei y p oblem mo e acu e. Ci il con lic will
p obably ha e an immedia e impac on people’s a i udes owa ds he i al g oups,
making hem mo e in ole an and pola ized and, hus, e e se causali y canno be
disca ded.
In he ollowing, we y o o e come his p oblem by ins umen ing he po en ially
endogenous eg esso s. Since, o he easons p o ided abo e, eligious F o RQ
indices could also be endogenous, we ocus on he e hnic dimension and ins umen
e hnic Pconside ing e hnic RQ measu es as exogeneous.
We ins umen e hnic P using a e ages o he dis ance be ween he languages spo-
ken in a coun y. Language dis ances a e a p oxy o he cul u al di e ences among
he g oups li ing in a e i o y. The iden i ica ion assump ion ha we adop is ha
language dis ances do no a ec con lic di ec ly bu only h ough i s co ela ion wi h
people’s ole ance o in ole ance o o he g oups. We belie e ha wha ma e s o
con lic is no he “objec i e" cul u al di e ences bu he way hese di e ences a e
pe cei ed and liked o disliked by he di e en g oups, dimension ha we a e able
o cap u e h ough ou pola iza ion indices. I his assump ion holds, a ia ions in
P induced by a e ages o language dis ances can be conside ed as exogenous and
employed o e alua e he e ec o an exogenous change in Pon he le el o con lic .
In addi ion, language-based indices a e e y s able o e ime and do no p esen he
e e se causali y p oblem ha po en ially a ec s e hnic P.
The e a e di e en ways o measu ing dis ances be ween languages. Fea on and
Lai in (1999, 2000) and Lai in (2000) p oposed o use he in o ma ion p o ided by
language ees. Language ees a e genealogical diag ams o languages ela ed by
descen o a common ances o . The dis ance be ween wo languages iand jis com-
pu ed as a unc ion o he numbe o common classi ica ions in he language ee. Fo
ins ance, Spanish and Basque di e ge a he i s b anch, since hey come om s uc-
u ally un ela ed language amilies. By con as , Spanish and Ca alan sha e hei i s
20
7 classi ica ions as Indo-Eu opean, I alic, Romance, I alo-Wes e n, Wes e n, Gallo-
Ibe ian and Ibe o-Romance languages. We ollow Fea on (2003) and Desme e al.
(2009) and de ine he dis ance be ween languages iand j o be
d
′
ij
= 1 −l
m
δ
,
whe e lis he numbe o common b anches be ween iand j,mis he maximum
numbe o sha ed b anches be ween any wo languages, and δis a pa ame e ha
de e mines how as he dis ance declines as he numbe o sha ed languages inc eases.
The weigh ed a e age o he dis ances be ween any wo pai o languages spoken in
a coun y is gi en by
WAD =
K

i=1
K

j=1
s
i
s
j
d
′
ij
,
whe e s
h
, h={i,j} deno es he sha e o people ha speaks language has a i s lan-
guage and Kis he o al numbe o languages spoken in a pa icula e i o y. We
will use WAD as ins umen o e hnic P.
16
Da a on WAD has been aken om Desme e al. (2009), who ha e elabo a ed
his a iable using he in o ma ion on language ees p o ided by he E hnologue
p ojec .
17
The dis inc i e cha ac e is ic o E hnologue e sus o he sou ces is i s
de ail. I p o ides e y disagg ega e in o ma ion o all he languages and dialec s
spoken in a e i o y. Fo ins ance, as no ed by Desme e al. (2009), he B i annica
Book o he yea 1990 edi ion epo s 21 li ing languages o Mexico while E hnologue
lis s 291. This means ha he sha es used o compu e WAD a e in gene al e y
di e en om he sha es employed o compu e P, since he WVS classi ica ion only
conside s a small numbe o ca ego ies. S ill, he co ela ion be ween e hnic P and
16
We ha e also conside ed o he indices o linguis ic dis ance, including he index o pola iza ion
o disc e e dis ibu ions P, as in (3). We ended up using he Gini-G eenbe g index ype WAD
because i ga e he highes co ela ion wi h he po en ially endogenous a iable.
17
Desme e al. (2009) choose a alue o δ= 0.05 o compu e d
ij.
21
WAD is 0.39.
Table 3 p esen s he esul s o ins umen ing e hnic P by WAD. As poin ed ou by
Ang is and K uege (2001), es ima ing a linea p obabili y model by wo s age leas
squa es (2SLS) is a obus es ima ion app oach e en i he unde lying second-s age
ela ionship is nonlinea , as in ou case.
The linea p ojec ion o e hnic P on he es o he con ols and WAD shows ha
he la e has a posi i e and highly signi ican e ec on e hnic P, wi h a p- alue o
0.019 and o 0.016, co esponding o he cases whe e e hnic RQ was included o no
in he i s -s age eg ession. Columns (1) o (4) p esen he esul s o ins umen ing
e hnic P in he disc e e dependen a iable eg ession while columns (5) and (6) ocus
on he con inuous indica o o con lic . Figu es in columns (1), (2), (5) and (6) ha e
been ob ained using 2SLS while hose in columns (3) and (4) wi h MLE in a p obi
speci ica ion.
The quali a i e esul s do no change wi h espec o hose epo ed in Tables 1
and 2: P
E h
is highly signi ican while RQ
E h
is no . Including o no RQ
E h
as an
addi ional eg esso does no ha e any impac on ei he he es ima ed coe icien o
P
E h
o on i s signi icance. Using he ule o humb sugges ed by Woold igde (2002),
i is possible o compa e he magni udes o he linea p obabili y and he p obi
speci ica ions. To do ha , we should di ide he p obi es ima es by 2.5, which u n
ou o be 1.46 and 1.44 o columns (3) and (4), espec i ely, and hus, he pa ial
e ec s implied by he p obi speci ica ion a e simila o hose o he linea p obabili y
model epo ed in columns (1) and (2). Using he la e igu es o e alua e he pa ial
e ec o e hnic pola iza ion on he p obabili y o incidence o ci il con lic , we ob ain
ha an inc ease in 1 s anda d de ia ion o P
E h
aises he p obabili y o ci il wa by
0.34.
18
18
O cou se, his implica ion canno li e ally be ue because con inually inc easing Pwould
e en ually d i e he p obabili y o con lic o be g ea e han one. Howe e , hese igu es can be
22
T 3. IV E!
Va iable (1) (2) (3) (4) (5) (6)
LGDPC −0.098
(0.029)
−0.095
(0.035)
−0.266
(0.107)
−0.261
(0.119)
−0.514
(0.000)
−0.516
(0.000)
LPOP 0.001
(0.982)
0.007
(0.814)
−0.003
(0.976)
0.011
(0.910)
0.119
(0.126)
0.113
(0.148)
PRIMEXP −0.291
(0.204)
−0.299
(0.223)
−1.342
(0.175)
−1.363
(0.173)
−1.195
(0.103)
−1.187
(0.105)
MOUNT 0.005
(0.119)
0.004
(0.144)
0.014
(0.088)
0.0121
(0.102)
0.006
(0.373)
0.008
(0.214)
NONCONT 0.120
(0.363)
0.136
(0.294)
0.522
(0.177)
0.565
(0.130)
0.823
(0.009)
0.797
(0.009)
DEM 0.013
(0.864)
0.010
(0.891)
−0.067
(0.784)
−0.075
(0.757)
0.134
(0.485)
0.141
(0.468)
RQ
E h
−0.217
(0.326)
-−0.431
(0.512)
-0.267
(0.541)
-
P
E h
1.241
(0.008)
1.190
(0.007)
3.660
(0.000)
3.614
(0.000)
1.749
(0.025)
1.808
(0.015)
Pseudo-R
2
- - - - 0.254 0.247
Coun ies 51 51 51 51 51 51
Obse a ions 385 385 385 385 344 344
No es
:
The dependen a iable PRIOCW (1)-(4) o ISC (5),(6).
The es ima ion me hod is 2SLS (1),(2),(5),(6) o IV P obi (3),(4). IV is WAD.
5.3. Robus ness o Resul s
The p e ious sec ion has shown ha including in e pe sonal dis ances in pola iza-
ion indices is ele an in explaining con lic and ha while RQ is no signi ican , P
is so in bo h he s anda d and he IV eg essions. In his sec ion we explo e he o-
bus ness o he p e ious esul s by conside ing al e na i e: a) sample, b) ins umen s,
c) de ini ions o ci il con lic , and d) inclusion o egional dummies. Fo he sake o
b ie ness we shall concen a e on he es ima es o e hnic pola iza ion.
good es ima es o he pa ial e ec s o Pnea he cen e o he dis ibu ion o he co a ia es.
23
D, J-Y., J. E,  D. R'
(2004), “Pola iza ion: Concep s, Measu e-
men , Es ima ion” Econome ica 72, 1737—1772.
E' W.  R. L
(1997), ÒA icaÕs G ow h T agedy: Policies and E hnic
Di isionsÓ, Qua e ly Jou nal o Economics 111, 1203-1250.
E, J., L. M'  D. R'
(2010), “E hnici y and Con lic : An Empi ical
S udyÓ, unpublished manusc ip .
E, J.  D. R' (1994), “On he Measu emen o Pola iza ion” Econo-
me ica 62, 819-852.
E, J.  D. R' (2008), “On he Salience o E hnic Con lic ," Ame ican
Economic Re iew 98, 2185-2202.
E, J.  D. R' (2010), “Linking Con lic o Inequali y and Pola iza ion,"
Ame ican Economic Re iew , o hcoming.
E, J.  G. S(2008) ÒPola iza ion and Con lic : Theo e ical
and Empi ical IssuesÓ (In oduc ion o he special issue), Jou nal o Peace Resea ch
45, 131-141.
F, J.(2003), “E hnic and Cul u al Di e si y by Coun y", Jou nal o Economic
G ow h 8, 195-222.
F, J.,  D. L (1999) “Weak S a es, Rough Te ain, and La ge-Scale
E hnic Violence since 1945”, P esen ed a he Annual Mee ings o he Ame ican
Poli ical Science Associa ion, A lan a, GA.
F, J.,  D. L (2000) “Violence and he Social Cons uc ion o E hnic
Iden i y,” In e na ional O ganiza ion 54, 845-877.
F, J.  D. L (2003), “E hnici y, Insu gency, and Ci il Wa ," Ame -
ican Poli ical Science Re iew 97, 75—90.
30

H, D.L. (1985), E hnic G oups in Con lic , Be keley, CA: Uni e si y o
Cali o nia P ess.
L, D. (2000) “Wha is a Language Communi y?,” Ame ican Jou nal o Poli ical
Science 44,142-155.
L, M. I. (1989), “An E alua ion o "Does Economic Inequali y B eed Po-
li ical Con lic ?", Wo ld Poli ics 4, 431- 470.
M, A. (2008) “His o ical S a is ics o he Wo ld Economy: 1-2008.AD",
h p://www.ggdc.ne /maddison/.
M-, M.I. (1988), “Rule s and he uled: pa e ned inequali y and he onse o
mass poli ical iolence,” Ame ican Poli ical Science Re iew 82, 491—509.
M*, E., S', S.  E. S* (2004), “Economic Shocks and Ci il Con-
lic : An Ins umen al Va iables App oach," Jou nal o Poli ical Economy 112, 725—753.
M, J. G.  M. R'-Q (2005), “E hnic Pola iza ion, Po en ial Con-
lic and Ci il Wa ," Ame ican Economic Re iew 95, 796—816.
M,E.N.  M.A. S* (1987) “Inequali y and Insu gency", Ame ican
Poli ical Science Re iew 81, 425-451.
M, E. N., M.A. S*  H. F (1989), “Land inequali y and poli ical
iolence", Ame ican Poli ical Science Re iew 83, 577—586.
N*, J. (1974), “Inequali y and discon en : a non-linea hypo hesis,” Wo ld Pol-
i ics 26, 453—472.
R'-Q, M. (2002), ”E hnici y, Poli ical Sys ems, and Ci il Wa s” Jou nal
o Con lic Resolu ion 46, 29-54.
S-, M. R., F. W. W'!  J. D. S* (2003) “In e -S a e, In a-
S a e, and Ex a-S a e Wa s: A Comp ehensi e Look a Thei Dis ibu ion o e
Time, 1816-1997," In e na ional S udies Qua e ly 47, 49-70.
31
S, G. N. W! (2006), “E hnic Pola iza ion, Po en ial
Con lic , and Ci il Wa s: Commen ", unpublished manusc ip , Uni e si y o Kon-
s anz.
W, M. C. (1994), “When Inequali ies Di e ge", Ame ican Economic Re iew
Pape s and P oceedings 84 353-358.
W*, J. M. (2002), Econome ic Analysis o C oss Sec ion and Panel
Da a. The MIT P ess.
32
Appendix A : The Cons uc ion o he Agg ega e Indices o A i udes
om he Wo ld Value Su eys
The Wo ld Value Su eys (WVS) p o ide since 1980´s an unique in e na ional da a on
he changing alues o he popula ion o di e en coun ies, including he opinions on he
issues ela ed o eligious and e hnic ole ance. We ha e used he ich WVS da a o y o
ack he in e -g oup dis ances ha a e needed o compu e he mul i-g oup DER.
The WVS a e conduced by means o ace- o- ace in e iews and includes hund e s o
ques ions asked in each pa icipa ing coun y. So a , he e ha e been conduc ed 5 wa es
o he WVS wi h pa icipa ion o 97 coun ies (a leas in one wa e). The coun y sample
a ies bu usually amoun s o 1000 esponden and is usually ep esen a i e wi h espec
o he age and sex s uc u e o each coun y and o en also s a i ied geog aphically. We
wish men ion he e he p oblem o he sampling ules ollowed by he WVS. Speci ically,
ha nei he eligion no e hnici y ha e been used o ob ain a balanced sample. Indeed,
he weigh o he di e en eligious g oups and o e hnici ies a ies signi ican ly om wa e
o wa e, p oducing an a i icial ime a iabili y o all he dis ibu ional indices. We ha e
op ed o aking he a e age o e all he wa es a ailable o each coun y.
The lis o ques ions a ies o e he wa es, and hence we ha e selec ed ques ions ha ha e
been asked in all he wa es o ques ions ha we e e y simila o e di e en wa es. In case
o eligious ole ance we use 7 and in case o e hnic 6 ques ions lis ed below. The ques ions
ha e ei he bina y o mul iple esponse wi h a mos 10 possible answe s ( ypically when
esponden is asked o ma k he opinion be ween wo opposi e s a emen s o he deg ee o
agg emen wi h ce ain s a emen ). We ake he aw su ey da a and ecode all he answe s
so ha 1 is he mos ole an one and 10 s ands o he mos in ole an one. I less han 10
answe s a e p o ided, we use he mean alues o he in e als, which numbe co esponds
o he numbe o possible esponses. Fo ins ance, o a ques ion "Do you belie e in God"
we assign alue o 3.25 o no and 7.75 o "yes" gi en ha yes can mean some hnig be ween
33
" o ally belie e" a he beli e han do no " and ise e sa o "no".
Consequen ly, we use a P incipal Componen Analysis (PCA) o collapse he esponse o
nume ous ques ions in o a single index numbe o each esponden (i.e. we e ain he i s
p incipal componen ). Compa ing densi y unc ions o di e en coun ies can p o ide some
e y basic e idence on he di e ences be ween eligious and e hnic eelings be ween di e en
coun ies. Howe e , he PCA p o ides by con uc ion a dis ibu ions wi h he same ze o
mean ha is no bounded. Fo he pu pose o ou analysis i is clea ly p e e able o wo k
wi h densi ies ha a e bounded on he same in e al o all he coun ies while he mean
o disc ibu ion di e . The e o e, we e ain he weigh s o he i s p incipal componen and
apply hem o he o iginal (no demeaned) alues o each a iable and bound he index on
he same in e al o he o iginal esponsed (1,10). To see his, no e ha egula PCA is
used o ind a ec o o weigh s ha maximize he a iance o a index zwhe e x
ij
is he
answe o pe son i o ques ion j.
z
i
=
j
α
j
(x
ij
−x
j
)
I we apply he ec o o weigh s on he o iginal a iables x, he dis ibu ion o he
esul ing index y does no ha e ze o mean anymo e bu i s scale a ies ac oss coun ies
gi en he sum o he weighs a ies ( he sum o he squa es o he weigh s is one):
y
i
=
j
α
j
x
ij
Howe e , we can he minimum and he maximum alue o he dis ibu ion o each coun y
so as o bound he index dis ibu ion on (1,10) in e al:
y
∗
i
=10 (y
i
−y
min
)
(y
max
−y
min
)
34
The abo e dis ibu ion bounded on (1, 10) can be simply compa ed o di e en coun ies.
Howe e , also o di e en g oups wi hin he same coun y. In he la e case, we use
a a iable ha iden ies he eligious g oup o e hnici y o esponden . Please no e, ha
iden i ica ion o he g oup o he esponden is necesa y so as o calcula e he DER mul i-
g oup pola iza ion index. This iden i ica ion is a ailable o 81 coun ies in he case o
eligion and 67 coun ies in case o e hnici y.
The ques ions used o ob ain he agg ega e indices a e he ollowing:
1. R* !
•V22 Religious ai h: do you conside i o be especially impo an ?
•V183. He e a e wo s a emen s which people some imes make when discussing
good and e il. Which one comes closes o you own poin o iew? A) The e
a e absolu ely clea guidelines abou wha is good and e il. These always apply
o e e yone, wha e e he ci cums ances. B) The e can ne e be absolu ely
clea guidelines abou wha is good and e il. Wha is good and e il depends
en i ely upon he ci cums ances a he ime.
•V185. Apa om weddings, une als and ch is enings, abou how o en do you
a end eligious se ices hese days?
•V186. Independen ly o whe he you go o chu ch o no , would you say you
a e: 1) a eligious pe son; 2) No a eligious pe son; o 3) a con inced a heis
•Which, i any, o he ollowing do you belie e in?
—V191 Do you belie e in God?
—V192 Do you belie e in li e a e dea h?
35

—V194 Do you belie e in hell?
—V195 Do you belie e in hea en?
•V196. How impo an is God in you li e? Please use his scale o indica e- 10
means e y impo an and 1 means no a all impo an .
•V197. Do you ind ha you ge com o and s eng h om eligion?
Gi en he ac ha ques ions V191, V192, V194, V195 a e qui e simila , we ha e
compu ed i s a e age alue as a single indica o .
2. E !
•On his lis a e a ious g oups o people. Could you please so ou any ha
you would no like o ha e as neighbo s?
—V69 People o a di e en ace
—V73 Immig an s/ o eign wo ke s
•V25 Gene ally speaking, would you say ha mos people can be us ed o ha
need o be e y ca e ul in dealing wi h people?
•V214 To which o hese geog aphical g oups would you say you belong i s o
all?
•V215 And he nex ?
—Locali y o own whe e you li e
—S a e o egion o coun y whe e you li e
Coun y as a whole
Con inen
36
—The wo ld as a whole
•V216 How p oud a e you o be [na ionali y]?
Appendix B: Desc ip ion o Da a
C   
Fo bo h con lic onse and incidence we use he a med con lic da ase om he
Upsala Con lic Da a P og am and he Peace Resea ch Ins i u e Oslo, UCDP/PRIO.
This da a se co e s om 1960 ill 2008, subjec o da a a ailabili y. Howe e , o keep
consis ency wi h MRQ da a ables 1 o 3 es ic o he pe iod 1960-1999 only. The
sample is di ided in o i e-yea pe iods. To eco d whe he a coun y is in con lic shall
ake he de ini ion o in e media e con lic . Acco dingly wi h PRIO2004 s anda ds
he de ini ion o In e media e con lic is “
PRIOCW
: mo e han 25 ba le- ela ed
dea hs pe yea and a o al con lic his o y o mo e han 1000 ba le- ela ed dea hs,
bu ewe han 1000 pe yea ." In ou obus ness checks we shall also wo k wi h he
no ions o “mino con lic " and “wa ". The co esponding de ini ions a e as ollows.
“
PRIO25
: be ween 25 and 999 ba le- ela ed dea hs in a gi en yea .",“
PRIO1000
: a
leas 1,000 ba le- ela ed dea hs in a gi en yea ."
C I8
This a iable is he con lic index compu ed by The C oss-Na ional Time-Se ies
Da a A chi e (CNTS). This index o con lic is he weigh ed a e age o eigh di e en
mani es a ions o domes ic con lic , adop ed om Rudolph J. Rummel, "Dimensions
o Con lic Beha io Wi hin and Be ween Na ions", Gene al Sys ems Yea book, VIII
[1963], 1-50).
20
The eigh a iables included a e:
20
The co ela ion wi h UCDP/PRIO is 0.45.W 9 PRIO 8?
37
•Assassina ions (domes ic1): Any poli ically mo i a ed mu de o a emp ed
mu de o a high go e nmen o icial o poli ician.
•Gene al S ikes (domes ic2): Any s ike o 1,000 o mo e indus ial o se ice
wo ke s ha in ol es mo e han one employe and ha is aimed a na ional
go e nmen policies o au ho i y.
•Gue illa Wa a e (domes ic3): Any a med ac i i y, sabo age, o bombings ca -
ied on by independen bands o ci izens o i egula o ces and aimed a he
o e h ow o he p esen egime.
•Majo Go e nmen C ises (domes ic4): Any apidly de eloping si ua ion ha
h ea ens o b ing he down all o he p esen egime - excluding si ua ions o
e ol aimed a such o e h ow.
•Pu ges (domes ic5): Any sys ema ic elimina ion by jailing o execu ion o po-
li ical opposi ion wi hin he anks o he egime o he opposi ion.
•Rio s (domes ic6): Any iolen demons a ion o clash o mo e han 100 ci izens
in ol ing he use o physical o ce.
•Re olu ions (domes ic7): Any illegal o o ced change in he op go e nmen
eli e, any a emp a such a change, o any success ul o unsuccess ul a med
ebellion whose aim is independence om he cen al go e nmen .
•An i-go e nmen Demons a ions (domes ic8): Any peace ul public ga he ing
o a leas 100 people o he p ima y pu pose o displaying o oicing hei
opposi ion o go e nmen policies o au ho i y, excluding demons a ions o a
dis inc ly an i- o eign na u e.
The weigh s used a e: Assassina ions (25), S ikes (20), Gue illa Wa a e (100),
Go e nmen C ises (20), Pu ges (20), Rio s (25), Re olu ions (150), and An i-Go e nmen
38
Demons a ions (10). The calcula ion is pe o med as ollows: weigh ed sum o oc-
cu ences o each e en di ided by 8 ( he numbe o ypes o e en s) and mul iplied
by 100.
I9 
We summa ize all he a iables used in ou empi ical exe cises in he ollowing
Table.
39