ethnic unmixing and civil war

This article was downloaded by: [University of South Florida]On: 08 October 2014, At: 07:11Publisher: RoutledgeInforma Ltd Registered in England and Wales Registered Number: 1072954Registered office: Mortimer House, 37-41 Mortimer Street, London W1T 3JH,UK

Security StudiesPublication details, including instructions forauthors and subscription information:http://www.tandfonline.com/loi/fsst20

Ethnic Unmixing and Civil WarDavid D. Laitin aa David D. Laitin is the James T. Watkins IV and EliseV. Watkins Professor of Political Science at StanfordUniversityPublished online: 25 Jan 2007.

To cite this article: David D. Laitin (2004) Ethnic Unmixing and Civil War, SecurityStudies, 13:4, 350-365, DOI: 10.1080/09636410490945938

To link to this article: http://dx.doi.org/10.1080/09636410490945938

PLEASE SCROLL DOWN FOR ARTICLE

Taylor & Francis makes every effort to ensure the accuracy of all theinformation (the “Content”) contained in the publications on our platform.However, Taylor & Francis, our agents, and our licensors make norepresentations or warranties whatsoever as to the accuracy, completeness,or suitability for any purpose of the Content. Any opinions and viewsexpressed in this publication are the opinions and views of the authors, andare not the views of or endorsed by Taylor & Francis. The accuracy of theContent should not be relied upon and should be independently verified withprimary sources of information. Taylor and Francis shall not be liable for anylosses, actions, claims, proceedings, demands, costs, expenses, damages,and other liabilities whatsoever or howsoever caused arising directly orindirectly in connection with, in relation to or arising out of the use of theContent.

This article may be used for research, teaching, and private study purposes.Any substantial or systematic reproduction, redistribution, reselling, loan,sub-licensing, systematic supply, or distribution in any form to anyone is

http://www.tandfonline.com/loi/fsst20

http://www.tandfonline.com/action/showCitFormats?doi=10.1080/09636410490945938

http://dx.doi.org/10.1080/09636410490945938

expressly forbidden. Terms & Conditions of access and use can be found athttp://www.tandfonline.com/page/terms-and-conditions

Dow

nloa

ded

by [

Uni

vers

ity o

f So

uth

Flor

ida]

at 0

7:11

08

Oct

ober

201

4

http://www.tandfonline.com/page/terms-and-conditions

ETHNIC UNMIXING AND CIVIL WAR

DAVID D. LAITIN

CHAIM KAUFMANN’S “Possible and Impossible Solutions to Ethnic CivilWars”1 makes the strongest case available that the best cure for ethnicwar is viable territorial defense through partition or substantial regional

autonomy, and that, in the wake of ethnic civil wars, attempts to bring secu-rity to intermingled ethnic groups by means other than partition or regionalautonomy are doomed to failure. The key premises of his analysis are, first,that hypernationalist rhetoric and atrocities are part and parcel of civil warand “harden ethnic identities.” This makes cross-ethnic appeals as a solutionto civil war likely to fail. Second, he takes as theoretically demonstrated that“intermingled population settlement patterns create real security dilemmasthat intensify violence, motivate ethnic ‘cleansing,’ and prevent de-escalationunless the groups are separated.” Therefore, he reasons, “stable solutions ofethnic civil wars are possible, but only when the opposing groups are demo-graphically separated into defensible enclaves.”2 As he develops his brief forthe separation of populations, Kaufmann systematically addresses several ob-jections to the solution he advocates. All solutions are bad, he concludes, butseparation is the least bad, even if least favored by policymakers.

The tight structure, extensive references, and tone of sang froid give this arti-cle an impressive cogency. Kaufmann’s concluding paragraph, urging that “wehave a responsibility to be honest with ourselves” not “to offer false hopes toendangered peoples,”3 leaves the reader with the feeling that cold logic and raw

David D. Laitin is the James T. Watkins IV and Elise V. Watkins Professor of Political Scienceat Stanford University.

This article presents data and arguments from a joint project with James D. Fearon. Supportfrom the National Science Foundation (Grants SES-9876477 and SES-9876530) and from theCarnegie Corporation of New York made the recoding of the group concentration variablesfor the MAR dataset possible. Ted Gurr and Victor Assal from the University of Marylandparticipated in the collection of these data, and kindly opened the MAR archives to Fearon andme. Matthew Kocher and Ebru Erdem directed the recoding operation. James Fearon, PeterKatzenstein, Roy Licklider, Jack Snyder, and Barbara Walter helpfully commented on earlierversions of this article.

1. Chaim Kaufmann, “Possible and Impossible Solutions to Ethnic Civil Wars,” InternationalSecurity 20, no. 4 (Spring 1966): 136–75.

2. Ibid., 137.3. Ibid., 175.

SECURITY STUDIES 13, no. 4 (summer 2004): 350–365Copyright C© Taylor & Francis Inc.DOI: 10.1080/09636410490945938

Dow

nloa

ded

by [

Uni

vers

ity o

f So

uth

Flor

ida]

at 0

7:11

08

Oct

ober

201

4

Ethnic Unmixing and Civil War 351

fact rather than dreamy delusions or wishful thinking motivate his recommen-dations. Nonetheless, there are two fatal flaws in his brief. The first concernshis inadequate rebuttal to the objection that secessionist warfare can encour-age further secession attempts elsewhere. Fearon (“Separatist Wars, Partition,and World Order,” in this volume) addresses this flaw. The second flaw con-cerns the issue of the security dimension of ethnic concentration. In this note,I address this second flaw. My objection does not undermine Kaufmann’sconclusion, which compares a variety of bad solutions. Rather, my objectionshould lead Kaufmann and policymakers considering the separation of pop-ulations to add substantially to Kaufmann’s assessment of its potential costs.

TERRITORIAL CONCENTRATION AND CIVIL WAR

IN HIS ANALYSIS of “demography and security dilemmas,”4 relying on Posen’swork,5 Kaufmann points out that the severity of ethnic security dilemmas

is greatest when demography is most intermixed, weakest when communitysettlements are most separate. The more mixed, or so the reasoning goes,the stronger the offense; the more separated, the stronger the defense. “Ac-cordingly [when mixed] each side has a strong incentive . . . to kill or driveout enemy populations before the enemy does the same to it, as well as tocreate homogeneous enclaves more practical to defend.” Ethnic militias inenclaves are less murderous, or so the reasoning goes, as they have defensivevalue only. Once groups are separated, “any attempt to seize more territoryrequires a conventional military offensive [and in consequence] normal deter-rence dynamics apply. Mutual deterrence does not guarantee that there will beno further violence, but it reduces the probability of outbreaks . . .”6

Kaufmann sees unmixing of ethnically different populations, once thesepopulations have been at war with each other, as the key to future interethnicpeace. He suggests that this can be achieved through partition (the creation ofde jure or de facto international boundaries separating the populations), or bygranting regional autonomy and demographic superiority to the minority pop-ulation in its own region.7 In this article, I do not address the consequences ofpartition for peace. Rather, I examine the consequences of enhanced regionalconcentration of ethnic groups within the boundaries of the state (throughthe granting of regional autonomy and the movement of peoples) for the

4. Ibid., 148.5. Barry R. Posen, “The Security Dilemma and Ethnic Conflict, Survival 35, no. 1 (Spring

1993): 27–47.6. Kaufmann, “Possible and Impossible Solutions to Ethnic Civil Wars,” 150.7. Ibid., 161.

Dow

nloa

ded

by [

Uni

vers

ity o

f So

uth

Flor

ida]

at 0

7:11

08

Oct

ober

201

4

352 SECURITY STUDIES 13, no. 4

likelihood of resurgent ethnic war. I present data, based on analysis of theMinorities at Risk (MAR) dataset,8 to assess the impact of political strategies toenhance regional ethnic concentration, and to challenge Kaufmann’s empiricalclaim that territorial concentration of ethnic groups results in relatively lowerprobability of violence even after civil wars.

The core finding from the analysis of MAR is that the concentration of groupswithin a region of a country, other things being equal, makes these groups bettercandidates for insurgencies leveled against states.9 In the MAR dataset, an eight-point scale of rebellion indicates the degree of violence in a civil war. Fearonand I have used the variable MAXREB60 to signify the maximum rebellionscore from 1960 through 1998.10 For group concentration, the MAR datasetincludes six different variables (five of them dummies) seeking to delineatethe spatial distribution of the group. GROUPCON is a summary index of the fivedummy variables. Fearon and I reported in a paper in 1999, referring to table 1below, that “being either “widely dispersed” or “primarily urban” (GROUPCON

= 0 or 1) proves to be almost a sufficient condition for a group to have a lowMAXREB score in these data.” We found that once per-capita GDP is controlledfor, GROUPCON is the only robust and significant predictor of group rebellionin the MAR dataset that is plausibly exogenous to the occurrence of grouprebellion (at least in so far as the data on concentration reflect populationpatterns before civil war outbreak).

Fearon and I have already reported several problems with the coding forGROUPCON, and our recoding effort to isolate better what was driving thisrelationship between group concentration and rebellion.11 The core conceptwhich we wished to clarify was that of a “regional base,” improving upon MAR’sconceptualization of group concentration. To determine whether a group hada regional base in a country, we first asked coders whether there were a spatiallycontiguous region larger than an urban area—that is, part of the country—in which a substantial fraction (25 percent or more) of the minority residedand in which the minority constituted the predominant proportion of the

8. The MAR dataset is available at www.bsos.umd.edu/cidcm/mar. For purposes of replica-tion of the models presented herein, request from the author marwork13.dta.

9. James D. Fearon and David D. Laitin, “Weak States, Rough Terrain, and Large-ScaleEthnic Violence since 1945” (paper presented at the annual meetings of the American PoliticalScience Association, Atlanta, GA, September 1999); Monica Toft, The Geography of Ethnic Violence(Princeton: Princeton University Press, 2003); Barbara Walter, “Explaining the Intractability ofTerritorial Conflict,” International Studies Review 5, no. 4 (December 2003): 137–53.

10. Scores are for five-year periods. The last period for which we have data ends in 1998.11. Fearon and Laitin, “Group Concentration and Civil War” (paper prepared for delivery

at the 2002 annual meeting of the American Political Science Association, Boston, 29 August–1 September 2002.

Dow

nloa

ded

by [

Uni

vers

ity o

f So

uth

Flor

ida]

at 0

7:11

08

Oct

ober

201

4


Table 1REGIONAL CONCENTRATION AND REBELLION IN THE MAR DATASET

GROUPCON MAXREB45≤3 ≥4 Total

Widely dispersed (Reg6)OR Primarily urban

(Reg5) OR minority inone region (Reg3) %

6788.2

911.8

76100

Majority in one region(Reg2 or Reg4) OR

Concentrated in oneregion (Reg1) %

10856.3

8443.7

192100

Total%

17565.3

9334.7

268100

χ 2d.f.=1 = 24.5, Pr = .000

Note: I assume a threshold of rebellion to be a score of ≥4 on the MAR rebellion variable.

population. If there were such a region, we asked coders to name that area. Weasked them also to assess whether this area was an administrative region of thestate (for example, Uzbek Soviet Socialist Republic in the Soviet Union) or atopographical region (for example, the Pyrenees), or a geographical expression(for example, the South).

We sought to capture two separate elements in our specification of a regionalbase. First, a reasonable percentage of the group (25 percent was our criterion)should live in a well-delineated region of the country. Second, the percentage ofthe country’s minority population living in that region ought to be significantlyhigher than the percentage of the country’s dominant group living in thatregion. In other words, to have a regional base, a substantial percentage ofthe minority group must be living in that region, and the probability that amember of this minority group is living in the region ought to be much higherthan the probability that a member of the country’s dominant group is livingin the region. We defined this formally in the expression below:

If:

mr ≡ minority population in region rmc ≡ minority population in country cMr ≡ majority population in region rMc ≡ majority population in country c

Dow

nloa

ded

by [

Uni

vers

ity o

f So

uth

Flor

ida]

at 0

7:11

08

Oct

ober

201

4


Table 2DESCRIPTIVE STATISTICS: WORLD REGION, REGIONAL BASE, AND REBELLION

World region

Percentage ofgroups withregional base

If regionalbase, percentin rebellionsince 1960

If no regionalbase, percentin rebellionsince 1960

Numberof groups

Western Democraciesand Japan

52 0 0 29

Eastern Europe andformer Soviet Union

64 27 0 58

Asia 37 84 23 59North Africa and

Middle East71 70 38 28

Sub-Saharan Africa 88 42 38 67Latin America and

the Caribbean59 26 0 32

Total 68.5 45 16 247

The percentage in rebellion is based on a score ≥4 on MAXREB60.

Definition: Minority m has a regional base in region r of country c if, andonly if, mr/mc � Mr/Mc and mr/mc > .25. (“�” means “is substantiallygreater than”).]12

As with GROUPCON, the regional concentration of groups—a dummy vari-able that asks whether a group has a “regional base” (GC2)—is a signifi-cant predictor of rebellion (here specified as whether the maximum rebellionscore for the eight coding periods since and including 1960—MAXREB60—ever reached the level of 4, our threshold of genuine rebellion).13 As indicatedin table 2, the probability of rebellion in all regions of the world save the West-ern Democracies and Japan (where no rebellions take place) is much greaterif the group has a regional base than if it does not. It is only in sub-Saharan

12. The Fearon/Laitin data for group concentration, that is, for groups having a regionalbase, differs from MAR’s composite variable GROUPCON, but the bivariate correlation is .64. Abreakdown of the differences is reported in Fearon and Laitin, “Group Concentration and CivilWar.” In this article (with one exception), for purposes of replication of Kaufmann’s reports,and despite problems of selection bias (addressed in the Fearon/Laitin NSF proposal noted inthe acknowledgements), I use only the 273 cases in the official MAR dataset.

13. I use rebellion scores since 1960 for the dependent variable because the data on popu-lation dispersion is based on 1960s estimates, and therefore the concentration variables cannotbe used to explain earlier rebellion levels (as reported in table 1).

Dow

nloa

ded

by [

Uni

vers

ity o

f So

uth

Flor

ida]

at 0

7:11

08

Oct

ober

201

4


Africa (where only 12 percent of the groups do not have a regional base, thusmaking the statistical description of likelihood of rebellion without civil warsomewhat skewed by the law of low numbers) where the probability of re-bellion comes even close for groups without a regional base. It is perhaps asign of the power of the regional base measure that the numbers are correctin Africa, even though four of the most rebellious groups—the Hutus andTutsis in Rwanda and Burundi—are widely dispersed.

This relationship between regional base and MAXREB60 holds up even whencontrolling for GDP and mountainous terrain, two robust predictors of rebel-lion in the country/year dataset.14 In the regression reported in table 3, Icontrol as well for ethnic fractionalization, which has come up as significant(or closely so) in several studies of rebellion. Here, it is also entered along withits squared version, which also comes out significantly in some country/yearanalyses.15 I also control for whether the group is in a sub-Saharan state whichwas a colony of France, a variable which has been significant in some of theFearon/Laitin models using the MAR dataset, and which approaches signifi-cance in this specification. Group concentration, even with these controls,comes out strongly. On the eight point scale of rebellion, other things beingequal, a concentrated group will have one-and-a-half times higher level in itsmaximum rebellion score since 1960 than a group which is not concentratedgeographically.

Kaufmann, to be sure, agrees about the “decisiveness of territory,” but histheory incorrectly leads him to predict that territorial control by one groupleads to a higher probability of peace. In fact, territorial control by one groupleads to a higher probability of war. He writes that, “population control de-pends wholly on territorial control. Since each side can recruit only for itsown community and only in friendly-controlled territory, incentives to seizeareas populated by co-ethnics are strong, as is the pressure to cleanse friendly-controlled territory of enemy ethnics . . .”; and he reasons, therefore, that“military control of the entire territory at issue is tantamount to total victory.”Thus, according to Kaufmann’s logic, once there are well-defined demographicfronts, “the strongest motive for attack disappears, since there are few or noendangered co-ethnics behind enemy lines.”16 The MAR data, as our discussion

14. I use OLS regressions here because they are easiest to interpret. When Fearon and Iincorporate these variables into more complex models, given the distribution on the dependentvariable, we will rely on different statistical techniques.

15. In Fearon and Laitin, “Ethnicity, Insurgency and Civil War,” American Political ScienceReview 97, no. 1 (February 2003): 75–90, neither ethnic fractionalization nor its square comesout significantly when regressed on civil war onset.

16. “Possible and Impossible Solutions to Ethnic Civil Wars,” 149.

Dow

nloa

ded

by [

Uni

vers

ity o

f So

uth

Flor

ida]

at 0

7:11

08

Oct

ober

201

4


Table 3THE INFLUENCE OF GROUP CONCENTRATION ON REBELLION

Dependent Variable: MAXREB60; OLS

[Specification: reg maxreb60 GC2 frassa ethfrac ethfrac2 lwbpop65 flat lgdp60enGC17r GC15all if (GC17r∼=6) & [marstat==31 | marstat==41 | marstat==53 |marstat==63 | marstat==73 | marstat==83]

Regional base (GC2) −1.59(.37)∗∗

Length of group’s residence in country (GC17r)(1) −.20(.15)

Transnational dispersion: kindred groups in power (GC15all) (2) −.067(.11)

GDP/Cap 1960, logged (Lgdp60en) −1.46(.22)∗∗

Flat terrain (flat) −1.32(.66)∗

Ethnic heterogeneity (ethfrac) −1.57(2.50)

Ethnic heterogeneity squared (ethfrac2) .74(2.65)

Country population (natural log) (lwbpop65) .19(.109)

French African colony (frassa) −1.35(.81)

Constant 13.45(2.34)∗∗

R-squared .29Number of observations 254

Notes for table 4: ∗ = significant at p < .05; ∗∗ = Significant at p < .01.(1) GC17r is GC17, but all Roma who were coded as GC17==6 were recoded as GC17==2,that is, they were coded as having arrived in their present country before the nineteenth century.The remaining cases of GC17==6 were deleted from the sample for this specification.(2) GC15all is GC15, but all cases in which GC14==1 (no close kindred across border) andthus get a missing value for GC15, are recoded as a 0 in GC15all, one step lower than having akindred group across the border with no access to power.

makes clear, do not support this analysis—groups that have some degree ofregional concentration appear to have stronger motives for attack comparedto groups which are more dispersed.

Kaufmann might offer two distinct rebuttals to the use of MAR data to chal-lenge his plea for partition or autonomy. He might point out that his theoryconcerns the phenomenon of “unmixing,” that is, of ethnic groups that had

Dow

nloa

ded

by [

Uni

vers

ity o

f So

uth

Flor

ida]

at 0

7:11

08

Oct

ober

201

4


been interspersed and subsequently, due to civil war, become concentrated.Thus—Kaufmann might offer as a conjecture—the process of fighting a war,agreeing to separation of the populations, and then actively moving into sepa-rate territorial zones could have a pacifying affect on the country. In contrast,as would be implied by this conjecture, concentration without previous war isdangerous. This is an intuitively perplexing position, since it would be difficultto explain why group concentration has such a violent effect on groups thathave not yet been at war, but a peaceful effect on groups that have ended awar. Nonetheless, it merits scrutiny.

Testing this conjecture, however, would require a clearer specification of theclaim. In most cases of ethnic war (as the MAR data make clear) there was al-ready concentration and certain degrees of regional autonomy before the war.Therefore to point out, as Kaufmann does, that successful resolutions to civilwars involve partition or autonomy is not fully informative.17 Such resolutionsdo not necessarily involve the movement of peoples or even changes in the ad-ministrative structure of the state.18 Kaufmann therefore needs better to spec-ify the criteria for unmixing—does it, for example, need to be greater than thestatus quo ante?—before one can test to see if his conjecture is correct. Thereare many examples of civil war endings in the past half-century—includingAbkhazia (Georgia), the IRA (Northern Ireland), the Tuaregs (Mali), Katanga(Congo-Kinshasa), and the Kurds (Turkey)—that have not involved substan-tial unmixing compared to the prewar situation. If the theory requires greaterautonomy or more unmixing than before the war, it is not clear that Kaufmann’sconjecture will withstand empirical scrutiny. Also, there are cases (Chechnyaand Southern Sudan) in which wars recurred after substantial grants of auton-omy. These cases decrease confidence in Kaufmann’s recommendation forautonomy.

A more systematic test of Kaufmann’s conjecture, relying on MAR data, isto isolate non-regionally concentrated groups that had been at war with theirstate, and in which the war ended. We could then ask whether autonomy andthe movement of peoples were necessary to avoid a recurrence? In the MAR

case list, there are only eleven groups that meet these criteria, and they are listedon table 4. Resolution in none of these cases involved autonomy along with thecompelled unmixing of populations. In six of the cases (55 percent) there was

17. Ibid., 160–61.18. Kaufmann at times combines partition and unmixing as having the same properties, but

at times separates them analytically. This leads to confusion. In “When All Else Fails: EthnicPopulation Transfers and Partitions in the Twentieth Century” (International Security 23, no. 2[Autumn 1998]: 120–56), for example, Kaufmann he writes of his four empirical cases “all. . . were accompanied by large-scale population transfers” (121). One of his cases, however, isIreland, which was a partition but which did not involve large-scale population transfers.

Dow

nloa

ded

by [

Uni

vers

ity o

f So

uth

Flor

ida]

at 0

7:11

08

Oct

ober

201

4


Table 4IS PARTITION AND THE UNMIXING OF POPULATIONS NECESSARY TO AVOID

RECURRENCE OF CIVIL WARS AMONG NON-CONCENTRATED GROUPS?

Group/country Rebel history Unmixing? Restart?

Santals/India 65–70 No (butquasi-partition)

No

Arabs/Israel 45 No NoHmong/Laos 45–75, 85– No (but attempt by

rebels to create anautonomous state)

Yes

Palestinians/Lebanon 65–90 No NoSunnis/Lebanon 55, 75–80 No Yes, NoBerbers/Morocco 55, 70 No Yes, NoHutus/Rwanda 60, 98 No YesTutsis/Rwanda 85–90 No NoBlack Africans/South Africa 75–80 No NoEast Pakistani Hindus/Pakistan

& Bangladesh45, 70 No Yes, No

Chinese/Malaysia 45–55 No No

Notes : (1) Cases include beginning of five-year period where maxreb45 > 3 (that is, there hasbeen at least one substantial rebellion) & GC2==2 (that is, the group was not concentratedregionally) and the rebellion after reaching >3, fell below 4 for at least one five-year period.(2) Rebel history (five-year periods in the MAR dataset when the rebellion score was >3).(3) In this table, I use all cases in the Fearon/Laitin dataset (see n. 12) to raise the number ofobservations for the rare phenomenon of civil wars for unmixed peoples. (The in-progressdataset, owing to an inappropriate coding rule, records the Chams in Cambodia and theSlovenes in Yugoslavia as not regionally concentrated. Although these cases, if included,would bias the results in my favor, I delete them from the analysis. Both groups at the time oftheir rebellions were geographically concentrated). The MAR dataset counts the East Pakistani(and Bangladeshi) Hindus as two observations—one in Pakistan and one in Bangladesh. Iconsolidate these into a single observation, as their mobilization in both eras was part of a singlehistory.

no recurrence of war, despite the lack of unmixing. Whites and Black Africansin South Africa have established a modus vivendi after a negotiated settlementthat did not involve autonomy or the substantial movement of peoples; Arabsinside the Green Line have not rebelled a second time after their initial rebellionin the wake of Israeli independence, and have lived in remarkable peace withJewish citizens in spite of regional chaos; the Chinese in Malaysia were at warfor over a decade, but today live in peace with the Malays without a regionalbase. The Santals (along with the Chotanagpurs, a related adivasi [tribal] group)live in the forests around the states of Madhya Pradesh, Orissa and Bihar. In

Dow

nloa

ded

by [

Uni

vers

ity o

f So

uth

Flor

ida]

at 0

7:11

08

Oct

ober

201

4


response to their insurgency, they were granted a provisional autonomouscouncil and in 2000 a new state of Jharkhand carved out only from Bihar. Thismight be counted as a legal separation, but it never involved the movement ofpeoples in or out of the autonomous area. Furthermore, scheduled tribes (ofwhich the Santals and Chotanagpurs are the predominant groups) constituteonly 28 percent of the population, which hardly reflects unmixing.19 This casetherefore does not fully meet Kaufmann’s criteria for full security. In threeof the cases (27 percent) there was recurrence, but there was subsequently asecond peace that has not broken, again with no unmixing. In one of thesecases, the East Pakistani and Bangladeshi Hindus, the recurrence was in thecontext of the Bangladeshi war of secession. With the success of the Bengalisecession, the Hindu rebellion was crushed. Hindus remain in Bangladesh inpeace and mixed with Bengali Muslims. These three cases have experienced arecurrence without unmixing, but they do not support Kaufmann’s predictionof hardened identities making it impossible to live next to one another in peace.After all, all three groups ultimately did. Finally, two cases (18 percent), theHutus in Rwanda and the Hmong in Laos, show a recurrence of rebellion afteran initial settlement that did not entail unmixing. These cases go as Kaufmann’stheory predicts. Kaufmann has already addressed the Hutu case.20 The othercase is the Hmong in Laos, who in 1966 (with aid from the Vietnamese)declared Meoland as their independent state. Pathet Lao forces crushed thisproto-state. It could be argued, in accordance with Kaufmann’s theory, that ifthe self-declared state had remained autonomous, there would not have beena war recurrence in the 1980s.

While there is no comparison set within MAR of post-civil war unmixingand substantial autonomy to compare results, it would be a violation of thehistorical record to claim that once civil wars begin among interspersed ethnicpopulations, it is not possible to resume civil peace without unmixing them andgiving each group an autonomous homeland. To the extent that it is possibleto specify Kaufmann’s theory to allow for a direct test using MAR data, thereis little support for it. Only two of the eleven cases lend it support.

19. Data are from the Jharkhand home page (jharkhand.nic.in/about/profile.htm).20. Kaufmann (“Possible and Impossible Solutions to Ethnic Civil Wars,” 168–69) makes

a sensible case for partition in Rwanda and Burundi. It is only an oddity of this “test” of histheory that the ongoing war of Hutus in Burundi is not included. Equally odd is the coding ofthe Tutsis of Rwanda as not having a recurrence, since this is due to the victory of the RPF, anarmy that they have dominated. In the MAR coding rules, a “minority” can capture the state,but if it does, it cannot have a rebellion against itself. The RPF, however, has shown that it candistinguish Hutus who perpetrated the genocide from those who did not; and there remainsa possibility of long term peace without unmixing. Kaufmann’s prediction, if Hutus return, of“sooner or later . . . another genocide” is wildly over-confident.

Dow

nloa

ded

by [

Uni

vers

ity o

f So

uth

Flor

ida]

at 0

7:11

08

Oct

ober

201

4


Kaufmann’s second rebuttal to the use of MAR data to test his theory mightbe that ethnic concentration as reported in the revised MAR dataset is insuf-ficiently informative about the ethnic demography. There could be macroconcentration (by the Fearon/Laitin coding for regional base) along with mi-cro mixing that will enhance the danger of the security dilemma. This toois an extraordinarily difficult issue to address empirically. In Posen’s classicarticle on the subject, he gives insufficiently precise criteria for judging thedegree to which a variety of ethnic demographies create a security dilemma.21

Consider in a rural area a situation in which two ethnic groups have 50 percentof the population but each lives in well-bounded villages that are themselveshomogeneous. Now consider a comparable case where each village has 50 per-cent of each group. It should be obvious that the security issues in these twodifferent regions are quite different, but they are lumped together in Posen’sarticle. Furthermore, cross-sectional data on this level of ethnic mixing are notavailable. It is therefore not possible with current data to decide conclusivelywhether this objection—that truth is in micro concentration—stands up toempirical scrutiny.

There is evidence, however, which should give pause to Kaufmann’s conclu-sion that ethnic mixing in the context of hostility is inherently unstable. Take,for example, Kaufmann’s data presented on his table 1.22 Here Kaufmann re-lies on Ted Gurr’s dataset on civil war endings in the past fifty years. From thistable, he concludes “the data supports (sic.) the argument that separation ofgroups is the key to ending ethnic civil wars.” This is an unfortunate misread-ing of at least two of the cases—and perhaps more of them. Kaufmann codesUkraine’s war against the USSR that began in 1945 as ending in de facto or dejure partition. He codes the Basque war vs. Spain beginning in 1959 as endingin autonomy. In neither of these cases, however, did the war settlement involvethe separation of groups. Russians continued to live peacefully in eastern andsouthern Ukraine after Ukraine became a federal republic. By 1989, 22 percentof the residents in the Ukrainian SSR were Russian.23 Extramadurans and othernon-Basques from Spain continued to live peacefully in Basque country afterautonomy was granted, and today make up about half the population of Paı́sVasco.24 Both civil wars ended without any significant separation of groups, indirect contradiction to Kaufmann’s assertion to the contrary. The “unmixing”

21. “The security Dilemma and Ethnic Conflict.”22. “Possible and Impossible Solutions to Ethnic Civil Wars,” 160.23. 1989 USSR Population Census, East View Publications, available on the web at www.

statistischedaten.de/products/ciscen.htm24. BBC news, 30 September 2002, news.bbc.co.uk/1/hi/world/europe/2289036.stm

Dow

nloa

ded

by [

Uni

vers

ity o

f So

uth

Flor

ida]

at 0

7:11

08

Oct

ober

201

4


of peoples, Kaufmann’s condition sine qua non, was unnecessary to preservethe peace at least in these two civil wars.

Other cases of ethnic wars in this list of ethnic civil wars resolved are notcogent examples of hardened identities making future mixing impossible. TheSomali clans have been at war for a decade and there is de facto partition;however, it would be a wild exaggeration to claim that the Isaaqs of thenow rump Somaliland Republic could never be secure in the south should apolitical solution be reached. Similarly in Ethiopia; while the Tigreans werevictorious in their war in Ethiopia, Amharas and Tigreans live peacefullyin Addis Ababa. Evidence does not support Kaufmann’s argument that theseparation of groups is the key to peace.

To be sure, ethnic war can have a powerful short-term effect of ethnic unmix-ing. Kaufmann provides illustrations of Ibos leaving the north in Nigeria in thewake of the massacres in Kano in 1966 and the subsequent civil war (1967–70)in which the federal government defeated the Biafran secession. Kaufmann,however, omits from his example the fact that after the war, Ibos returned inlarge numbers to Kano, the site of the massacres. Indeed it would be hard toaccount for the killings and intimidation of Ibos in Kano and other northerncities connected with the Sharia court crisis of 1990 had there been a full sep-aration of groups in the wake of the Biafran civil war.25 Similarly in Sri Lanka.Kaufmann points to the rapid unmixing in Colombo after the massacres in1983. He omits mentioning, however, that shortly thereafter the population ofTamils in Colombo continued to increase. In 1953, 9.5 percent of the popula-tion in Colombo was Tamil. In 1981, the figure rose to 11.2 percent. Supposethere was a vast exodus in 1983. By the time of the 2001 census, the percent-age of Tamils in Colombo had increased to 12.4 percent.26 If civil wars andmassacres make ethnic mixing unbearable, it is difficult to explain why mem-bers of formerly massacred groups would return to the scene of the crimes.Kaufmann’s evidence of fleeing ethnics with “hardened . . . identities,” unableto live in security mixed with the ethnic other, is undermined by the demo-graphic facts of voluntary return. Most likely, middle class Tamils considered

25. The Nigerian census of 1991 did not include questions of ethnicity and religion,and it therefore is not possible to get precise figures on Ibo populations in Kano. SeeRotimi T. Suberu, Federalism and Ethnic Conflict in Nigeria (Washington, D.C.: U.S. Insti-tute for Peace, 2001), 157. On the flight of Ibos in 2000 from northern cities, con-sult www.usafricaonline.com/shariashowdown chido.html, a report from 29 February 2000.Kaufmann (“When All Else Fails,” 156) addresses the security implications of Ibo return fornon-Ibos, but doesn’t address the question of why, if antipathies are so hardened after civil wars,Ibos returned to the sites of the 1966 massacre directed against them.

26. See Census of Ceylon 1953 (Columbo: Department of Census and Statistics, 1957), sec. 4,table 16. For 1981 and 2001, see www.statistics.gov.lk/abstract/population/ch02tab10.htmand www.statistics.gov.lk/abstract/population/ch02tab09.htm. Thanks to Kanchan Chandrafor these data and to Val Daniels for help in the analysis of them.

Dow

nloa

ded

by [

Uni

vers

ity o

f So

uth

Flor

ida]

at 0

7:11

08

Oct

ober

201

4


Table 5ETHNIC MIXING IN REGIONAL BASE AND REBELLION

Population of group/total population Mean of maximum rebellionliving in regional base (GC7a) scores since 1960 (maxreb60)

25–50% 2.7051–75% 4.3276–100% 3.31

Note : Only groups with regional base are included. Of the 187 groups with a regional base, 102were missing values (coders not able to make a reasonable estimate).

Colombo (where they would not face armed confrontations) safer than Jaffna.In Colombo, through the creation of micro security zones with fellow Tamilsand links to officialdom in the army and police, Tamils secure a modicumof protection. However secured, their return demonstrates that political au-tonomy and mass population exchange is not the only hope for peace in SriLanka.

Two final pieces of evidence put to question the argument purporting to linkethnic mixing, security dilemmas and ethnic war. These concern the proportionof the minority group in the region and in the country. First, consider table 5, inwhich I present the mean maximum rebellion score for groups that constitutedifferent percentages of the total percentage in their regional base. ConfirmingKaufmann’s point, the highest maximum rebellion scores are for groups thatare most mixed in regional bases (50–75 percent).27 The greatest security,however, comes from having the regional group more insecure in its region(25–50 percent of the population) than more secure (75–100 percent). Fromthese ambiguous data, it would be foolhardy to take as strong a position asKaufmann has done in regard to the relationship of ethnic mixing and theprobability of renewed ethnic war.

Second, the MAR dataset has information on the proportion of the groupin the country population. If I add that to the regression presented on table 3,the result is a near significant positive relationship with rebellion.28 In onesense, this confirms Kaufmann’s security dilemma argument inasmuch as there

27. The difference between the average maximum rebellion scores for the 50–75 percentand 75–100 percent groups, however, is not statistically significant in a two-tailed t -test (p =.11). I am using a measure of “group percent of regional base” that is measured for 1960 andis missing many observations (a total of 100 groups). When I use the same variable measuredfor 1990, which has 133 observations, the average rebellion scores, in order of increasing grouppercentage in region, are 2.1, 4.1, 3.6

28. I added the variable “gpro98,” which is the proportion of the group in the country in1998. It would be inappropriate to present this in the regression on table 3 because data from

Dow

nloa

ded

by [

Uni

vers

ity o

f So

uth

Flor

ida]

at 0

7:11

08

Oct

ober

201

4


should not be a security dilemma if the group is too small for any hope forsuccess in fighting. In fact, smaller groups are less likely to rebel. In anothersense, the data do not work well with Kaufmann’s overall argument, as theyimply that giving large groups their own region will make them even moredangerous than if they were dispersed.

This accumulating but insufficiently decisive evidence debunking the argu-ment about ethnic mixing leads to further reflections on another observableimplication of Kaufmann’s security dilemma argument. If the argument iscorrect, we should see the effects of regional concentration to be stronglypositive on rebellion when states are weak than when they are strong. If statesare strong, they can protect minorities, and therefore mixing should be no lesssecure than separation. If states are weak, however, minorities are less likelyto be protected. Under these conditions, regional concentration should serveas a substitute for state protection.

The data give little support to this observable implication. To test it, I dividedthe groups up into two halves—each group had a GDP value for its country. Icomputed the median group/country/GDP value to separate the groups as towhether they were living in rich or poor countries. The assumption here is thatcountry wealth is a proxy for state capacity to provide order. In both rich andpoor states, the mean rebellion score goes substantially down under conditionsof the group not having a regional base. (For groups in states with greater thanor equal to median GDP, maximum rebellion score goes from .78 for groupswith no regional base to 2.18 for groups with a regional base. For groupsin states with less than the median GDP, the maximum rebellion score goesfrom 1.61 for groups with no regional base to 3.96 for groups with a regionalbase.) Rebellion is less likely with ethnic mixing in both strong and weakstates. This finding suggests that regional concentration is not a substitute forstate protection for ethnic minorities, again challenging the security dilemmaframework. This test is not perfect. Rich countries may have the capacity butnot the desire to offer protection to regionally concentrated minorities. Evenso, it reduces confidence in a theory of civil war resting on a security dilemmafoundation.

Why, then, is the situation of unmixed populations which do not face asecurity dilemma so dangerous for rebellion in the past half century, whenKaufmann’s theory predicts the opposite? For one, Kaufmann adds additional

1998 cannot be used to explain rebellion from 1960. If, however, there were available groupproportion data from 1960, I doubt the result would be much different. In the run with gpro98,the coefficient is 2.17 and the standard error is 1.15. No significant changes occur for the othervariables in the model.

Dow

nloa

ded

by [

Uni

vers

ity o

f So

uth

Flor

ida]

at 0

7:11

08

Oct

ober

201

4


criteria for the peacefulness of regional bases, and does not base his entiretheory on settlement patterns. While the separation of populations need notprovide sovereignty to afford security, he judges, it ought to have a “regionalself-defense capability that abrogating the autonomy of any region would bemore costly than any possible motive for doing so” (p. 162). Also, he argues,peace cannot be maintained if there are militarily significant minorities in anyregion. It could well be the case that the regional bases in the MAR datasetconsistently lack these criteria, and that is why they have been so dangerous.

There are, however, other avenues to explore which may account for the MAR

data that contradict Kaufmann’s theory. Consider Kaufmann’s assumption thatgroups having their own territory lose their strongest motive for attack.29 Thisneeds to be reassessed for two reasons. First, ideologues in groups that livein a regional base sometimes have visions of a homeland far larger than theircurrent base, and have ambitions for fulfilling an historic mission of reachingthose boundaries. In our coding of concentration patterns, ideologues in 96out of the 171 MAR groups (56 percent) have articulated a vision of theirhomeland that far exceeds their current regional base.

An example of this phenomenon may be culled from Kaufmann’s case stud-ies of partition, in which he presents evidence that in the wake of violence,partition has an ameliorating affect.30 Post-partition ideas of a wider homeland(wider than was granted in the partition) by Israelis drove the government toencourage pioneers to settle in secure Arab zones, undermining both commu-nities’ security. The reality of territorial control—whether through partitionor regional autonomy—along with a dream to incorporate a wider homeland,give groups an incentive to disrupt the peace. Kaufmann argues that much ofthe violence in places such as Punjab and Ireland come from the “incomplete-ness of separation,”31 but the incommensurate boundaries of regional baseand imagined homeland make incompleteness—at least in cases of contestedimaginary homelands—a virtual certainty.32

Second, without sovereignty, groups in militarily secure regions can still betaxed, and from their point of view, unfairly so. Ibos, for example, were notafraid of extermination if they remained in the eastern region of Nigeria in1966, but they fought to retain rights to eastern region oil revenues.33 So any

29. “Possible and Impossible Solutions to Ethnic Civil Wars,” 162.30. “When All Else Fails.”31. Ibid., 121.32. Thus, given sharp and conflicting views between Muslims and Hindus about the bound-

aries of their homeland, I am less certain than Kaufmann (“When All Else Fails,” 141–42) thatKashmir could have been cleanly partitioned.

33. Micahel Ross, “How Does Natural Resource Wealth Influence Civil War? Evidence from13 Case Studies,” International Organization 58 (Winter 2004): 35–67.

Dow

nloa

ded

by [

Uni

vers

ity o

f So

uth

Flor

ida]

at 0

7:11

08

Oct

ober

201

4


granting of autonomy can lead to further demands for sovereignty, re-ignitingviolent conflict, especially if the state faces lost mineral reserves if sovereigntywere granted.

On a larger scale, Fearon and I have suggested an alternative theory toKaufmann’s. The importance of group concentration as the only group levelvariable that clearly and consistently comes out significantly in predictingrebellion—while variables such as economic inequality, religious discrimina-tion, and denial of recognition of a group’s ancestral language, are either not sig-nificant or not robust across a range of model specifications—provides strongsupport for the interpretation provided in Fearon’s and my country/year anal-ysis.34 The conditions allowing for successful insurgency, we argue, are a betterguide to the occurrence of civil violence than are the social, economic, politicaland cultural conditions that might motivate a group to organize a rebellion.Consistent with this interpretation, the concentration of a group in a countrycan be seen as allowing for low-cost communication among the populationin whose name the insurgency is being fought. In our reckoning, concentra-tion gives insurgents a strategic advantage in knowing who amongst them iscollaborating with the state, and the concomitant ability to restrain and sanc-tion such collaboration. Thus a regionally concentrated group raises the costsfor the government to weed out insurgents from a densely planted minoritypopulation. Insurgents are activated, we argue as an alternative to Kaufmann,with the strategic advantages for warfare that group concentration allows.

I have not made here a cost assessment of alternate solutions to civil war ascompared to partition or autonomy. Indeed Kaufmann (“When All Else Fails”)provides several examples where partition has lowered violence, and so hisbrief cannot be ruled out entirely. I have shown, however, that autonomy alongwith the ethnic unmixing of peoples, the least bad outcome in Kaufmann’sordering, has a dire implication ignored in his analysis. I have also shown thatthere have been successful solutions to civil wars fought by non concentratedgroups that did not involve concentrating them. Kaufmann pleads that allsolutions other than separation of populations provide only “false hopes.”As it currently stands, his plea overestimates the risk of post-civil war ethnicmixing and underestimates the risks of partition or autonomy coupled withunmixing. The hope he offers is no more true, as the evidence now stands,than the ones he calls false.

34. “Ethnicity, Insurgency and Civil War.”

Dow

nloa

ded

by [

Uni

vers

ity o

f So

uth

Flor

ida]

at 0

7:11

08

Oct

ober

201

4

ethnic unmixing and civil war

Documents