No Lessons Learned from the Holocaust?

Assessing Risks of Genocide and Political Mass Murder

since 1955
Author(s): Barbara Harff
Source: The American Political Science Review, Vol. 97, No. 1 (Feb., 2003), pp. 57-73
Published by: American Political Science Association
Stable URL: http://www.jstor.org/stable/3118221
Accessed: 07-01-2016 05:38 UTC

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 [email protected].

American Political Science Association and Cambridge University Press are collaborating with JSTOR to digitize, preserve
and extend access to The American Political Science Review.

http://www.jstor.org

This content downloaded from 146.189.194.69 on Thu, 07 Jan 2016 05:38:52 UTC
All use subject to JSTOR Terms and Conditions
American Political Science Review Vol. 97, No. 1 February2003

NoLessonsLearnedfromthe Holocaust?AssessingRisks
of Genocide and Political Mass Murder since 1955
BARBARA HARFF u.s. Naval Academy
his article reports a test of a structuralmodel of the antecedentsof genocide and politicide (political
mass murder). A case-control research design is used to test alternative specifications of a mul-
tivariate model that identifies preconditions of geno-/politicide. The universe of analysis consists
of 126 instances of internal war and regime collapse that began between 1955 and 1997, as identified
by the State Failure project. Geno-/politicides began during 35 of these episodes of state failure. The
analytic question is which factors distinguish the 35 episodes that led to geno-/politicides from those that
did not. The case-control method is used to estimate the effects of theoretically specified domestic and
international risk factors measured one year prior to the onset of geno-/politicide. The optimal model
includes six factors thatjointly make it possible to distinguish with 74% accuracy between internal wars
and regime collapses that do and those that do not lead to geno-/politicide. The conclusion uses the model
to assess the risks of future episodes in 25 countries.

W W e must remember not only what hap- recent decades genocide and Holocaust scholars have
pened but why and how it happened." joined to combat ignorance about the causes of these
In remembering the Holocaust "we need events, to engage politicians, and to mobilize pub-
to learn its lessons and apply them to contemporary lic empathy. Some have listened. The Clinton Ad-
events" (Mattas 1992, 185). What has been learned? ministration, in the aftermath of Rwanda, sponsored
We know that genocides and political mass murders the use of social science analysis to explain geno-
are recurrent phenomena; that since WWII nearly 50 cides and ethnic conflict, with an eye to developing
such events have happened; that these episodes have early warning systems to detect humanitarian disas-
cost the lives of at least 12 million and as many as 22 ters in the making. The study reported here was sup-
million noncombatants, more than all victims of inter- ported in part by two successive administrations and
nal and international wars since 1945;1and that human builds on years of prior research by those involved in
suffering rarely mobilized policymakers into action. the comparative study of genocide and similar pheno-
We also know that despite Cambodia, Bosnia, and mena.
Rwanda, some lessons have been learned. During During the early 1980s social scientists began to
study genocide comparatively (see Fein 1979, 1992;
Harff 1987,1992; Kuper 1981; and Melson 1992). These
Barbara Harff is Professor, Department of Political Science, U.S.
Naval Academy, Annapolis, MD 21401 ([email protected]).
pioneers, surveyed in Totten and Jacobs 2002, defined
This study was commissioned in 1998 by the Central Intelli-
the phenomenon, accumulated evidence of past and
gence Agency's Directorate of Intelligence in response to President ongoing cases, and went on to develop explanations for
Clinton's policy initiative on genocide early warning and prevention. the occurrence of such events (see Fein 1993b, chap. 3,
It was designed by the author and carried out using her data with for a review of theoretical approaches). Comparative
other data and analytic techniques developed by the State Failure
Task Force. Statistical analyses reported here were done by Michael genocide research has drawn upon conflict analysis.
Lustik and Alan N. Unger of Science Applications International Cor- Although a distinct phenomenon, it shares some char-
poration (SAIC), McLean, Virginia. The author is senior consultant acteristics and antecedents with ethnic wars and rev-
to the Task Force, which was established in 1994 in response to a olutions. Moreover, almost all genocides of the last
request from senior U.S. policymakers to design and carry out a data- half-century occurred during or in the immediate after-
driven study of the correlates of state failure, defined to include rev-
math of internal wars, revolutions, and regime collapse.
olutionary and ethnic wars, adverse or disruptive regime transitions,
and genocides and politicides (for the latest report on Task Force Ideologies that mobilize potential revolutionaries can
research see Goldstone et al. 2002). The author acknowledges the also incite ethnic hatred and provide incentives to kill
advice of other Task Force consultants and analysts throughout the real or perceived enemies of the new order. However,
research process. The views expressed here are those of the author
and do not represent the official views of the U.S. government, the despite the explosive growth of the literature, only two
U.S. intelligence community, or the Central Intelligence Agency. published genocide studies systematically test various
The author especially thanks Ted Robert Gurr for his critiquing hypotheses derived from the case and comparative lit-
early drafts and using the findings to construct the table that identifies erature (Fein 1993a; Krain 1997).
high-risk countries and groups. His insistence about the importance This article expands on previous theoretical and em-
of the study prompted me to revise the manuscript a number of
times, despite my initial reluctance, given the years of work that had pirical work by testing the effects of prior conflict,
gone into its preparation. It was especially hard to condense this
elite characteristics,regime type, and international con-
effort from its original 75 pages. The paper also benefited from a text on the likelihood of geno-/politicide. The opti-
careful reading by Mark I. Lichbach of a previous report and from mum model identifies six preconditions of genocide and
comments of anonymous reviewers for the American Political Science
Review. politicide (political mass murder) that make it possible,
1 Numbers of geno-/politicides and fatalities are tabulated for the list using the case-control procedure and logistic regres-
of cases and fatalities in Table 1, with the addition of episodes that sion, to postdict accurately 74% of episodes that began
occurred between 1945 and 1954 reported in Harff 1992. between 1955 and the late 1990s.

57
This content downloaded from 146.189.194.69 on Thu, 07 Jan 2016 05:38:52 UTC
All use subject to JSTOR Terms and Conditions
 No
No Lessons Learned from the
the Holocaust?February
Lessons Learnedfrom
Holocaust? 2003
February2003~~~~~~~~~

DEFININGAND IDENTIFYINGEPISODES tional definition of genocide and politicide used here

OF GENOCIDE AND POLITICIDE is expanded to include episodes that occur during civil
wars when a territorially based nationalist or revolu-
From Legal to Empirical Definition tionary movement targets an ethnic or political group
for destruction "in whole or in part." This encom-
Genocide, according to Article II of the United Nations
passes situations in which at least one party to a civil
(UN) Genocide Convention, refers to "acts committed war systematically uses deadly force to destroy the civil-
with intent to destroy, in whole or part, a national, eth- ian support base of its opponents, as in Angola's civil
nical, racial, or religious group." Points b, c, d, and e wars since the early 1970s.
specifically refer to conditions whose cumulative ef- The following definition summarizes the above
fects are conductive to a group's destruction. These
points and is used to identify the universe of cases
points are (b) "causing serious bodily or mental harm for comparative analysis. Genocides and politicides are
to members of the group,"(c) "deliberately inflicting on the promotion, execution, and/or implied consent of
the group conditions of life calculated to bring about its sustained policies by governing elites or their agents-
physical destruction in whole or in part," (d) "imposing or, in the case of civil war, either of the contending
measures intended to prevent births within the group," authorities-that are intended to destroy, in whole or
and (e) "forcibly transferring children of the group to
another group." This legal definition provides the basis part, a communal, political, or politicized ethnic group.
In genocides the victimized groups are defined by their
for an operational definition but has four limitations.
perpetrators primarilyin terms of their communal char-
First, the Convention does not include groups of acteristics. In politicides, in contrast, groups are defined
victims defined by their political position or actions.
primarily in terms of their political opposition to the
Raphael Lemkin (1944) coined the term genocide and regime and dominant groups. In common usage the
later sought the support of as many states as possible for Kurds of Iraq are said to be victims of genocide. In
a legal document that would outlaw mass killings and fact many Iraqi Kurds serve in the Iraqi bureaucracy
prescribe sanctions against potential perpetrators. Be- and military and some are members of the ruling Baath
cause the first draft of the Convention, which included
Party. The Kurds who were targeted for destruction in
political groups, was rejected by the USSR and its al- the al Anfal campaign of 1987 were the mainly rural
lies, the final draft omitted any reference to political
mass murder (Le Blanc 1988). The concept of politi- supporters of the Kurdish Democratic Party and the
Patriotic Union of Kurdistan. Thus, the event was a
cide is used here to encompass cases with politically
defined victims, consistent with Fein's (1993b, 12) line politicide (see Makiya 1992).
The definition parallels those developed by other
of reasoning that "mass killings of political groups show
similarities in their causes, organization and motives." comparative researchers. For example, Fein (1993b, 24)
exhaustively reviewed definitional discussions and pro-
Second, the phrase "mental harm" in point b is prob- posed a sociological definition: "Genocide is sustained
lematic, for it encompasses a vast array of instances of purposeful action by a perpetrator to physically de-
psychological and cultural harm done to groups that stroy a collectively directly or indirectly, through in-
have lost their cohesion and identity, but not their lives, terdiction of the biological and social reproduction of
as a result of processes of social and economic change.
group members, sustained regardless of the surrender
Setting aside this problematic clause, the crime of geno- or lack of threat offered by the victim."2Her definition
cide is delimited to acts that collectively endanger the differs from the one used here mainly in its lack of
physical life of group members. reference to the identity of the perpetrators, i.e., states
Third, the phrase "intent to destroy" raises the ques- or rival authorities (for a similar definition see Chalk
tion of how observers can reliably infer the intentions of and Jonassohn 1990, 23).
authorities. Perpetrators rarely signal their intentions
as clearly as Hutu extremists did in Rwanda in the early
1990s when they characterized Tutsis as vermin and Operational Guidelines
proposed to eradicate them (see Des Forges 1999 and In genocides and politicides killings are never acciden-
Prunier 1995, chaps. 5, 6). The operational guidelines
used here to infer intent and apply them to specific cases tal, nor are they acts of individuals. The key is that
are described below. they are carried out at the explicit or tacit direction
of state authorities, or those who claim state authority.
Fourth, the Genocide Convention does not take into The following guidelines were used to help distinguish
account the possibility that nonstate actors can and do
cases of genocide and politicide from other kinds of
attempt to destroy rival ethnic and political groups. One
killings that occur during civil conflicts. (1) Is there com-
unambiguous example from the post-Cold War period
is the ethnic cleansing perpetrated by Serb nationalists plicity by the state (or, in the case of civil war, either
of the contending authorities) in actions undertaken
against Muslims in Bosnia (1992-93). A possible case that endanger human life? (2) Is there evidence, even
occurred in Congo (Kinshasa) in late 1996 and early
if circumstantial, of intent on the part of authorities to
1997. Suspicions are strong that, during and after the
isolate or single out group members for mistreatment?
Kabila-led revolution that overthrew Mobutu's gov-
ernment, Kabila knew of and probably endorsed the
systematic killings of Hutu refugees in eastern Congo 2 For a detailed
analysis of alternative social science definitions and
by Tutsi members of his army. Therefore the opera- episodes encompassed by them, see Fein 1993b, 8-31, 79-91.

58
This content downloaded from 146.189.194.69 on Thu, 07 Jan 2016 05:38:52 UTC
All use subject to JSTOR Terms and Conditions
American Political Science
American Political Science Review
Review Vol.
Vol. 97, No.
No. 1

(3) Are victims members of an identifiable group? (4) can politicides, friends and relatives of leftist activists
Are there policies and practices that cause prolonged often were killed even though they themselves were
mass suffering? and (5) Do the actions committed pose politically inactive.
a threat to the survival of the group?
Threat to Group Survival. It is wrong to assume that
Establishing the Complicity and Intent ofAuthorities. most or all members of a group have to be eliminated
Any persistent, coherent pattern of action by the state before one can conclude that a genocide occurred. It
and its agents (or, in the case of civil war, either of is enough to "take the life out of the group"-in other
the contending authorities) that brings about the de- words, to eliminate or intend to eliminate so many peo-
struction of a collectivity, in whole or part, is prima ple that the group ceases to function as a social or po-
facie evidence of authorities' responsibility. Note that I litical entity. Thus, in politicides perpetrators typically
do not presume that only states can commit genocide. attempt to destroy the ability of opposition groups to
The Kabila-led revolutionary movement in Congo is a challenge or resist the regime by targeting their poten-
recent case in point. In most historical cases, however, tial supporters. Again, this point is closely related to
the state or its agents have been the perpetrators. intent. It follows that, in principle, "body counts" do
A related issue concerns the duration of a group's not enter the definition of what constitutes an episode.
victimization. The physical destruction of a people re- If an authority's motive is to rid itself of unwanted op-
quires persistent, coherent actions authorized by those position by destroying a group, and if policies with that
in power. Brief episodes of killings are not counted. intent are sustained over a substantial period of time,
Thus isolated massacres are not included in the list of then a few hundred deaths constitute as much a geno-
episodes, such as massacres of Palestinians in Beirut's cide or politicide as the deaths of tens of thousands.
Chatilla and Sabra refugee camps in September 1982. For example, about 900 Iranian Baha'is were victims
The shortest episode on the list is Rwanda, where the of genocide, as defined above, during the Khomeini
killing campaign in spring 1994 lasted about 100 days. regime.
The Question of Intent. Human rights scholars fre-
quently argue that there is no need to include intent Cases of Geno-/Politicide from 1955 to 2001
among the necessary conditions that lead one to con- The general definition and operational rules guided the
clude that a genocide is in the making. On the contrary,
I think that it is important to look for evidence that compilation and successive revisions of a list of geno-
allows us to infer intent precisely so that genocide cides and politicides since World War II. This list is
can be distinguished from related phenomena. More- widely accepted by researchers, several of whom have
used it for comparative research, sometimes adding or
over, early warning efforts depend on detecting sig-
nals of intent rather than waiting for information that deleting a few cases.3 The 37 cases cited in this analysis
include all those that began after 1955. Several consid-
widespread killings have taken place. How do we detect erations led to the 1955 starting point. Most episodes
intent?
in the late 1940s and early 1950s were continuations
* Potential perpetrators are agents of the state or rival of prior conflicts, e.g., four cases in the USSR that fol-
authorities, for example, military or police units, or lowed through on Stalin's wartime campaigns against
militias authorized by the state or by revolutionary disloyal national peoples and potential dissidents. As
leaders. a consequence of decolonization, many new, conflict-
* Elites and groups linked to them often use hate pro- prone states entered the international system beginning
paganda and attack ethnic and political opponents of in the 1950s, and as a practical matter, reliable data for
the state. most independent variables were sparse or nonexistent
* Government repression in response to opposition ac- before then.
tivities is greatly disproportionate to opposition acts. A full list of cases used for the analysis appears in
* Authorities and security forces ignore isolated Table 1. Three less-known cases are described briefly
killings and abuse of ethnic and political group below, emphasizing the information supporting the in-
members. ference that authorities intended the target groups' de-
struction.
Rare is a situation such as Nazi Germany, in which
Hitler's Mein Kampf clearly advocated elimination of Guatemala from July 1978. Revolutionary conflict
a people. Pol Pot comes close, in the sense that Khmer began in mid-1966. The military-linked death squads
Rouge ideology explicitly identified its future victims. responsible for killing leftists were clearly govern-
ment sponsored. Their actions became systematic and
Identity of Victims. The question is whether the vic-
tims belong to an identifiable ethnic, religious, or po- widespread, rather than episodic, after General Lucas
Garcia became president in July 1978. The Carter Ad-
litical group, either a self-defined collectivity or one
ministration was well aware of government complicity
authoritatively defined as such. In some cases victims in the killings and imposed restrictions on U.S. mili-
may not see themselves as members of a group but have
been ascribed characteristics that led to their victimiza- tary aid because of human rights violations. The most
tion. In Nazi Germany, people who changed their reli-
gion from Judaism to Christianity were still identified 3
Comparative empirical studies that use this list of episodes include
and targeted for elimination as Jews. In Latin Ameri- Fein 1993a, Krain 1997, Licklider 1995, and Schmeidl 1997.

59
This content downloaded from 146.189.194.69 on Thu, 07 Jan 2016 05:38:52 UTC
All use subject to JSTOR Terms and Conditions
No Lessons Learned from the Holocaust? February2003
February

TABLE 1. Genocides and Politicides from 1955 to 2001

Countryand Dates Natureof Episode EstimatedNumberof Victims
Sudan, 10/56-3/72 Politicidewithcommunalvictims 400,000-600,000
South Vietnam,1/65-4/75 Politicide 400,000-500,000
China,3/59-12/59 Genocide and politicide 65,000
Iraq,6/63-3/75 Politicidewithcommunalvictims 30,000-60,000
Algeria,7/62-12/62 Politicide 9,000-30,000
Rwanda,12/63-6/64 Politicidewithcommunalvictims 12,000-20,000
Congo-K,2/64-1/65 Politicide 1,000-10,000
Burundi,10/65-12/73 Politicidewithcommunalvictims 140,000
Indonesia,11/65-7/66 Genocide and politicide 500,000-1,000,000
China,5/66-3/75 Politicide 400,000-850,000
Guatemala,7/78-12/96 Politicideand genocide 60,000-200,000
Pakistan,3/71-12/71 Politicidewithcommunalvictims 1,000,000-3,000,000
Uganda,2/72-4/79 Politicideand genocide 50,000-400,000
Philippines,9/72-6/76 Politicidewithcommunalvictims 60,000
Pakistan,2/73-7/77 Politicidewithcommunalvictims 5,000-10,000
Chile,9/73-12/76 Politicide 5,000-10,000
Angola, 11/75-2001 Politicideby UNITAand governmentforces 500,000
Cambodia,4/75-1/79 Politicideand genocide 1,900,000-3,500,000
Indonesia,12/75-7/92 Politicidewithcommunalvictims 100,000-200,000
Argentina,3/76-12/80 Politicide 9,000-20,000
Ethiopia,7/76-12/79 Politicide 10,000
Congo-K,3/77-12/79 Politicidewithcommunalvictims 3,000-4,000
Afghanistan,4/78-4/92 Politicide 1,800,000
Burma,1/78-12/78 Genocide 5,000
El. Salvador,1/80-12/89 Politicide 40,000-60,000
Uganda, 12/80-1/86 Politicideand genocide 200,000-500,000
Syria,4/81-2/82 Politicide 5,000-30,000
Iran,6/81-12/92 Politicideand genocide 10,000-20,000
Sudan, 9/83-present Politicidewithcommunalvictims 2,000,000
Iraq,3/88-6/91 Politicidewithcommunalvictims 180,000
Somalia,5/88-1/91 Politicidewithcommunalvictims 15,000-50,000
Burundi,1988 Genocide 5,000-20,000
Sri Lanka,9/89-1/90 Politicide 13,000-30,000
Bosnia, 5/92-11/95 Genocide 225,000
Burundi,10/93-5/94 Genocide 50,000
Rwanda,4/94-7/94 Genocide 500,000-1,000,000
Serbia, 12/98-7/99 Politicidewithcommunalvictims 10,000
Note: This list of episodes was compiledin a long-termresearcheffort(see Harff1992), has been updatedand modifiedforthe State
FailureTaskForce,and is posted on the Universityof Maryland'sCenterforInternational Developmentand ConflictManagementWeb
platform,http://www.bsos.umd.edu/cidcm/inscr/stfail.The list and analysis reportedhere exclude a few episodes identifiedin previous
studies, for example, in Angola in 1961-62 (because it was then a colony), in EquatorialGuinea in 1969-79 (the countrywas below
the 500,000 populationthresholdused in the State Failureanalyses), in Paraguayagainstthe Ache Indiansin 1962-72, and in Nigeria
against Iboslivinginthe Northin 1966 (inthe lattertwo cases the governmentwas not complicitin killingscarriedout by privategroups).
Estimatesof victimsare invariably impreciseand oftenvarywidelyamongscholars,journalists,humanrightsobservers,and spokesmen
forthe victimizedgroups.Some of the figuresare littlemorethan guesses. Ifa detailedand reliablestudy is available,a single figureis
used. A single figureis also used when several sources offersimilarestimates. Whendifferentestimates are reportedand there is no
basis forchoosing among them, a range is shown.

massive rights violations, including multiple massacres and militia. From 1983 to 1985 tens of thousands of
and forcible resettlement of Mayan villagers, were car- Bugandan sympathizers of rebel commander Museveni
ried out by military units whose identities have been were killed in the "Luwero triangle" north of Kampala
publicly identified (a comprehensive bibliography is (see Kasozi 1994 and Minority Rights Group 1984).4
given in Ball, Kubrick, and Spirer 1999, 135-54; see
also Quigley 1999).
4 There were two distinct
episodes of genocide in Uganda, each with
Uganda After December 1980. The genocide perpe- different perpetrators and victims. But Amin's genocide ended with
trated by General Idi Amin's government from 1972 his overthrow in April 1979, and the onset of Obote's reprisals in
to early 1979 is well known. Less known are events late 1980 began less than two years later. For purposes of statistical
that followed Milton Obote's reinstallation as presi- analysis (below) the events are treated as a single case beginning in
dent in December 1980. His government encouraged February 1972. Similarly in Pakistan, two episodes-against Bengali
nationalists in breakaway Bangladesh in 1971 and rebellious Baluchi
massive reprisal killings against ethnic groups that were tribesmen in 1973-were separated by less than two years and are
regarded as loyal to Amin. Killings were carried out by treated in statistical analysis as a single case beginning in January
groups allied to Obote as well as by government troops 1971.

60
This content downloaded from 146.189.194.69 on Thu, 07 Jan 2016 05:38:52 UTC
All use subject to JSTOR Terms and Conditions
American
American Political Science Review
Political Science Review Vol. 97, No.
Vol. 97, No. 1

Sri Lanka After July 1989. This is one of the very rare Cambodian case, or retributive, as in Iraq.5 The sce-
contemporary instances of mass murder carried out in nario that leads to an ideological genocide begins when
a democratic state. The Marxist Sinhalese JVP (Janatha a new elite comes to power, usually through civil war
Vimukthi Perumuna) organized a revolutionary cam- or revolution, with a transforming vision of a new soci-
paign against the Sinhalese-dominated government in ety purified of unwanted or threatening elements. Fein
1971 (which was suppressed) and began again in 1987- (1984, 18) characterizes such exclusionary ideologies
89. By mid-1989 they were close to military victory as "hegemonic myths identifying the victims as out-
and rejected government efforts to negotiate a settle- side the sanctioned universe of obligation." The sta-
ment. Death squads, made up of military and police and tus of the victims is evident in the labels applied to
working at night, were authorized in September 1988 to them, such as "class enemies," "counterrevolutionar-
track and execute JVP members but had limited use un- ies," and "heretics." Case studies suggest that the more
til July 1989, when the by-now-desperate government intense the prior struggle for power and the greater
gave the military a free hand to do anything neces- the perceived threat the excluded group poses to the
sary to eliminate the JVP. Death squad killings rapidly new regime, the more likely they are to become victims
increased, the targets being Sinhalese youths through- of geno-/politicide. Examples of ideologically inspired
out the poor rural areas where most JVP supporters geno-/politicide occurred in China during the Cultural
lived. The leader of the JVP was captured and killed in Revolution of 1966-75, in Marxist Ethiopia in the late
November and by January 1990 the JVP hierarchy had 1970s, in Chile after the overthrow of the leftist Allende
been eliminated, its support base terrorized into passiv- regime in 1973, and in Iran after the 1981 revolution.
ity, and the killings stopped (Gunaratna 1990; Human Whereas ideological genocides can be thought of as
Rights Watch 1991). outcomes of elite succession struggles, retributivegeno-/
Although the last half-century has seen a long-term politicides are strategies forged during and in the im-
increase in numbers of geno-/politicides, by the end mediate aftermath of civil wars. Some occur during a
of 2001 mass killings were ongoing in only two coun- protracted internal war-whether ethnic or revolution-
tries, Sudan and Angola, fewer than at any time in the ary or both-when one party, usually the government,
previous 30 years. As Table 1 shows, several countries systematically seeks to destroy its opponent's support
were responsible for multiple cases. In Burundi the base. Examples occurred in South Vietnam in the 1960s
Hutus were victimized in three separate episodes; the and early 1970s; during both phases of the Sudanese
Rwandan Tutsis, the Iraqi Kurds, and Southerners in civil war; in East Timor after 1975; and in Guatemala,
Sudan each were targeted twice. Five other countries, Angola, and Sri Lanka. Some retributive episodes oc-
China, Indonesia, Pakistan, Uganda, and the Demo- cur after a rebel challenge has been militarily defeated.
cratic Republic of Congo, had two episodes in which In Indonesia in 1965-66, for example, a coup attempt
different groups were victimized. Because some coun- supposedly inspired by Communists led to countrywide
tries had multiple episodes, the analysis addresses massacres of party members and other civilians.
methodological concerns about the statistical interde- Since geno-/politicides almost always occur in the
pendence of cases by estimating alternative models. context of violent political conflict and regime change,
can they be explained in the framework of theories of
PRECONDITIONS FOR GENOCIDE AND revolutionary or ethnic conflict, or of political insta-
POLITICIDE: VARIABLES AND INDICATORS bility in general? Not directly, because the puzzle ad-
dressed here is why some such conflicts lead to episodes
Building and testing a general model of the precon- of mass murder, whereas most do not. Theories of the
ditions for genocide and politicide assume a coherent etiology of civil conflict, e.g., Goldstone 1991 on revo-
universe of analysis. The answer is outlined in the previ- lutions, Horowitz 1985 on ethnic conflict, and Tilly 1978
ous section: The 37 cases are deliberate and sustained on political mobilization, do not address this question.6
efforts by authorities aimed a destroying a collectiv- More relevant are analyses of the causes of state re-
ity in whole or in part. The theoretical objective is to pression and maintenance of social order (see Lichbach
identify general conditions under which governments, 1987,1998). The approach taken here focuses on factors
and rival authorities in internal wars, choose such a that affect the decision calculus of authorities in conflict
strategy. Do genocides and politicides need separate situations, in particular, the circumstances that lead to
theoretical explanations? Many episodes combine el- decisions to eliminate rather than accommodate rival
ements of both. The "Cambodian genocide" was both groups. Several scholars propose that the greater the
a politicide, because the Khmer Rouge targeted class threat posed by challengers, the greater the likelihood
and political enemies, and a genocide, because they
also targeted urban Chinese and the Muslim Cham. 5
The terms come from Fein (1984, 8-22), who also distinguishes
The Iraqi regime's 1987 al Anfal campaign aimed to two other historically common types of genocide. Despotic episodes
eliminate rural Kurds in rebellious areas but not all were aimed at peoples who resisted the imposition of colonial or
Kurds; thus, it was a politicide against some members imperial rule. Developmental genocides targeted indigenous peoples
of an ethnic group. The model tested here combines all for elimination when they resisted settlement and exploitation of
their lands.
cases, although others may choose to test whether the 6 There is no index entry for genocide in any of these books. Nor
genocide/politicide distinction leads to different results. is there any such entry in a 1980 edited volume that surveys the
Almost all genocides and politicides of the last half- empirical literature on violent civil and international conflict (Gurr
century were either ideological, exemplified by the 1980).

61
This content downloaded from 146.189.194.69 on Thu, 07 Jan 2016 05:38:52 UTC
All use subject to JSTOR Terms and Conditions
No Lessons Learned from the Holocaust? February
February2003

that regimes will choose massive repression (Fein 1993; The key variable in the analysis is the magnitude of
Gurr 1986; Harff 1987). They also argue for the impor- political upheaval; the argument is that the greater the
tance of habituation: Successful uses of violence to seize extent of violent conflict and adverse regime change,
or maintain power establish agencies and dispositions the greater the likelihood that geno-/politicide will oc-
to rely on repression in future conflicts. Krain (1997, cur. There are two rationales. First, the more intense
332-37), among others, proposes that regime changes, and persisting conflict has been, the more threatened
including those that follow revolutions, open up op- authorities are likely to be, and the more willing to take
portunities for elites to eliminate groups that might extreme measures. Second, following Krain (1997), the
challenge them. Numerous researchers point out that greater the extent of political disruption, the greater the
democratic norms and political structures constrain opportunities for authorities to seek a "final solution"
elite decisions about the use of repression against their to present and potential future challenges.
citizens whereas autocratic elites are not so constrained Annual magnitudes of each episode of state failure
(Gurr 1986; Henderson 1991; Rummel 1995). Others were estimated via ordinal ratings of the scope and in-
suggest that the international environment is a major tensity of violent conflict and its impact on governance.8
source of both incentives for and constraints on the The indicator of political upheaval is the sum of magni-
elite's use of repression. Thus a country's peripheral tudes of all state failures occurring in the 15 years pre-
status in the international system can be a permissive ceding the onset of geno-/politicide. Conceptually, the
condition (Harff 1987), and international war can pro- 15-year specification was intended to capture the extent
vide a cover (Melson 1992), whereas international en- of disruption due to long-term conflicts; empirically,
gagement and condemnation can constrain repression. the 15-year aggregation gave somewhat better analytic
Each of the general factors, below, is interpreted in results than 10- and 20-year aggregations. Although it
terms of its likely effects on authorities' choices about may be that revolutions, ethnic wars, and regime col-
whether to resort to mass killings in conflict situations. lapse have different effects on the likelihood of geno-/
politicide, because the purpose here is to estimate an
efficient model for a relatively rare kind of event, the
Political Upheaval: The Necessary
Precondition for Genocide and Politicide aggregate concept and indicator are used.

The beginning point ispolitical upheaval, a concept that Prior Genocides: Habituation to Mass
captures the essence of the structuralcrises and societal Killings?
pressures that are preconditions for authorities' efforts
to eliminate entire groups (Harff 1987; Melson 1992). Recall that ten countries had multiple episodes of geno-
Political upheaval is defined as an abrupt change in the / politicide in the last 45 years. Events that began four or
political community caused by the formation of a state more years after the end of a prior event were treated as
or regime through violent conflict, redrawing of state separate cases in the analyses reported here. I am mind-
boundaries, or defeat in international war. Types of po- ful of Fein's (1993a) observation that perpetrators of
litical upheaval include defeat in international war, rev- genocide often are repeat offenders, because elites and
olutions, anticolonial rebellions, separatist wars, coups, security forces may become habituated to mass killing
and regime transitions that result in the ascendancy of as a strategic response to challenges to state security
political elites who embrace extremist ideologies. and, also, because targeted groups are rarely destroyed
Empirically, all but one of the 37 genocides and in their entirety. To test for "repeat offender" effects, a
politicides that began between 1955 and 1998 occurred binary indicator of whether or not a previous genocide
during or immediately after political upheavals, as de- had occurred in the country since 1955 was used.
termined from the State Failure's roster of ethnic and
revolutionary wars and adverse regime changes:7 24 Political Systems: Exclusionary Ideologies
coincided with ethnic wars, 14 coincided with revolu- and Autocratic Rule
tionary wars, and 14 followed the occurrence of adverse
regime changes. As these numbers imply, several geno-/ Political upheaval is a necessary but not sufficient con-
politicides began after multiple state failures of differ- dition for geno-/politicide. Between 1955 and 1996
ent types. In addition, four of the 37 sequences of state more than 90 state failures did not lead to geno-/
failure followed the establishment of an independent politicide within four years of their onset. Two charac-
state, either through decolonization or breakup of an teristics of political governance have vital intervening
existing state. effects-the ideological commitments of elites and the
extent of democratic constraints on their actions.
7 The State Failure project's lists of ethnic and revolutionary wars Contending elites usually have many strategic and
and adverse regime changes are posted at http://www.cidcm.umd. tactical options for defeating or neutralizing opposi-
edu/inscr/stfail. The one geno-/politicide that did not coincide tion groups. Elite ideologies are crucial determinants
with a "failure" is the 1981-82 Syrian case in which the Muslim of their choices. Episodes of genocide and politicide
Brotherhood was targeted because its revolutionary objectives were become more likely when the leaders of regimes and
a threat to the regime. The Brotherhood's rebellion is below the
threshold of revolutionary was used in the State Failure data set.
8 The scales used to assess magnitudes of each type of conflict and
Since it fits the larger pattern of retributive geno-/politicides, it
is included in the count of episodes that coincided with revolu- the annual scores assigned to each failure event are posted on the
tionary war. web site given in footnote 7.

62
This content downloaded from 146.189.194.69 on Thu, 07 Jan 2016 05:38:52 UTC
All use subject to JSTOR Terms and Conditions
American PoliticalScience
American Political ScienceReview
Review Vol. 97,
Vol. No. 1
97, No.

revolutionarymovements articulate an exclusionary rights and the inclusion of political opponents,while

ideology, a belief system that identifiessome overrid- competitiveelections minimizethe chancesfor adher-
ing purposeor principlethatjustifieseffortsto restrict, ents of exclusionaryideologiesto be elected to highof-
persecute,or eliminatecertaincategoriesof people. In fice.Thisis the domesticequivalentof the "democratic
effect, an exclusionaryideology increases elites' op- peace" argument (see Davenport 1999 and Hegre,
portunitiesto eliminate groups, includingthose that Ellingsen,Gates,and Gleditsch2001).Thereare,how-
pose no obvious threat to the elite-like the Cham ever, historicaland contemporaryinstancesof geno-/
who were targetedby the KhmerRouge. Racism and politicide being carried out by governmentswith a
antisemitismare historicalexamples.Fein (1993a,95) semblanceof democraticinstitutions.Ideologies that
observesthat "ideologiesof exclusionare now masked excluded indigenous peoples from the universe of
by new goals:socialistutopias,cultsof personality,reli- obligation and justified their destruction coexisted
gious messianism,exclusivenationalisms.... Ideology with democratic institutions for white majorities in
is most apt to be a motivatingfactor duringand after nineteenth-centuryAustralia and the United States.
revolutionaryupheavalsor transfersof power...." SriLanka'sdemocraticgovernmentcarriedout a politi-
Elites with "exclusionaryideologies" were opera- cide againstJVP supportersat a time when the gov-
tionallydefinedto includethe following.9 erningelite was seriouslychallenged.These exceptions
* Adherents of strict variantsof Marxism-Leninism, aside,the generalpropositionis that a state that main-
tains democraticgovernancein the face of state fail-
as in the German Democratic Republic through- ure is muchless likely to commitgeno-/politicidethan
out its history.Laos, Vietnam,the People's Repub- autocraticregimes.
lic of China (PRC), and North Korea all are coded
as Marxist-Leninistthrough 1998. State socialist Autocracyand democracywere indexed using the
Polity global data set's 0- to-10 point scales based on
regimes that tolerated some civil society organiza- codedinformationon politicalinstitutions(Jaggersand
tions and/or allowed significantfree enterpriseare Gurr1995).Fulldemocracieshavea democracy-minus-
not coded as "exclusionary"; examplesare Hungary
afterthe 1960sand Polandthroughoutthe period of autocracyscoreof 7 to 10,andpartialdemocracieshave
a score of from 1 to 7; i.e., they have some democratic
Communistrule. features but lack others. In autocraciescitizens' par-
* Rulers of Islamic states governed on the basis of
Shari'alaw,as in Iran,SaudiArabia,and Sudan.Is- ticipationis sharplyrestricted;chief executivesare se-
lectedwithinthe politicalelite;and,once in office,chief
lamic states such as Bahrainand Oman that permit executivesexercisepower with few or no institutional
some expressionof other religionsare not coded as constraints.Full democraciesrarelyfail, and thus have
exclusionary. virtuallyno risk of geno-/politicide,whereas partial
* Advocates of rigid anticommunistdoctrines such
as those advanced by military-dominatedelites in democracies,especiallypoor ones, have a substantially
Taiwanand South Korea until the 1980s and "na- higherriskof failurethanautocracies(Esty et al. 1999,
tional security"regimes in some Latin American 52-55; Goldstone et al. 2002).10The question in this
countriesduringthe 1960sand1970s(see Lopez 1986 studyis whetherfailed states with partialdemocracies
havebeen less likelythanautocraticregimesto commit
for an analysisof nationalsecurityideologyandiden-
tificationof regimesmotivatedby it). geno-/politicidessince 1955.
* Advocates of doctrinesof ethnicand ethnonational-
ist superiorityor exclusivity,as in Iraq,SouthAfrica Ethnic and Religious Cleavages
duringApartheid,Serbia,and Bhutan.In Indonesia Ethnic and religious divisions are often identified as
the Suhartoregime'sdoctrineof "PancasilaDemoc-
racy"also was treatedas an exclusionaryideology. preconditionsof civilconflictin general(Rabushkaand
* Advocates of doctrinesof strictsecularnationalism Shepsle1972),ethnicconflict(Gurr2000;Kuper1977),
thatexcludepoliticalparticipationof religiousmove- and geno-/politicidespecifically(Chalkand Jonassohn
ments, as in Turkey,Egypt, and Algeria (note that 1990;Kuper1981).I showed above that violent ethnic
Turkeyand Egypt exclude only movementsthat ex-
plicitly use Islamicsymbols in their advertisements 10
Updated and validated Polity data are posted at http://www.
or partyplatforms). bsos.umd.edu/cidcm/inscr/polity. Other indicators of democracy and
autocracy have been developed (see Munch and Verkuilen 2002) but
Democraticand quasi-democraticregimeshave in- lack, in varying degrees, the conceptual distinctions and temporal
stitutionalchecks on executive power that constrain and geographic coverage of Polity. Validation studies by the State
elites from carryingout deadly attackson citizens,as Failure Task Force used Arthur S. Banks's updated time-series data
noted above.Moreover,the democraticnormsof most on variables such as legislative effectiveness (coding categories in
Banks 1971) separately and jointly with the Polity data (including
contemporarysocietiesfavorthe protectionof minority individual dimensions of Polity such as constraints on the chief ex-
ecutive) to assess the effects of various aspects of democracy and
autocracy on various types of state failure. The Task Force also ex-
9 This and other indicators described below are included in the State perimented with alternative cutting points on the Polity scales to
Failure Task Force's data dictionary and data set posted at the website assess the susceptibility of full and partial democracies and full and
given in footnote 7. The categories were developed and applied by partial autocracies to various types of state failure (see Goldstone
the author, using information in standard political handbooks; the et al. 2002). A number of these alternatives were used for estimating
coding was later replicated by an independent team of coders and the geno-/politicide model. Results using the most efficient predictor
discrepancies were resolved by reference to additional sources. are reported here.

63
This content downloaded from 146.189.194.69 on Thu, 07 Jan 2016 05:38:52 UTC
All use subject to JSTOR Terms and Conditions
No Lessons Learned from the Holocaust? February2003
February

conflict was a precursor to nearly two-thirds (24 of 37) ethnicity is not salient; 1 = elite ethnicity is salient-
of the geno-/politicides of the last half-century. But the political leadership is representative of the largest
what are the most direct linkages? One possibility is communal group or a coalition of several groups that
sheer diversity:The greater is ethnic and religious diver- together constitute a majority; and 2 = elite ethnicity is
sity, the greater the likelihood that communal identity salient-the political leadership is representative of a
will lead to mobilization and, if conflict is protracted, minority communal group or a coalition of small groups
prompt elite decisions to eliminate the group basis of that together constitute less than a majority.
actual or potential challenges. A plausible alternative
is that small minorities in otherwise homogeneous soci-
eties are at risk, especially when a regime is committed Low Economic Development
to an exclusionary ideology.11 To test these and simi- The State Failure project has consistently found that
lar arguments, indicators of diversity were constructed armed conflicts and adverse regime changes are more
from a cultural data set that records the size of ethnic,
likely to occur in poor countries. This is true glob-
linguistic, and religious groups for each state from 1820 ally and for sub-Saharan Africa, Muslim countries, and
through the 1990s, compiled by Philip Shafer of the countries with ethnic wars (Goldstone et al. 2002). The
Correlates of War (COW) project. fact that most geno-/politicides of the last 50 years oc-
A second and more complex connection begins with curred in Africa and Asia suggests that a similar re-
differential treatment of some groups. Discrimination lationship should be included in this structural model.
against a communal (religious or ethnic) minority is The hypothesis is tested using an indicator of infant
likely to increase the salience of group identity and its mortality: reported deaths to infants under one year
mobilization for political action (Gurr 2000, 66-72). per 1,000 live births. Extensive validation analyses by
If political action takes the form of persistent com- the Task Force suggest that infant mortality rates are
munal rebellion, regimes are likely to respond with a good surrogate for a wide range of indicators of
repression, which, when other predisposing structural material standard of living and quality of life. They
variables are present, can escalate into campaigns to have greater reliability and give somewhat better em-
eliminate the group. A country-level binary indicator pirical results in models of risks of state failure than
of active discrimination was constructed for the State indicators of Gross Domestic Product (GDP)/Gross
Failure project from group-level data collected by the National Product (GNP).
Minorities at Risk project: If any group in a country was
subject to active political or economic discrimination in
a given year, the country received a positive score. International Context: Economic
A third possible connection is the ethnic and reli- and Political Interdependence
gious composition of the political elite. If the elite dis- International context matters for geno-/politicides.
proportionately represents one segment in a heteroge-
neous society, two consequences follow that may lead Shifting global alliances like those that followed the end
of the Cold War decreased the predictability of interna-
to genocidal outcomes. Underrepresented groups are
tional responses to instability and gross human rights
likely to challenge the elite's unrepresentativeness, and violations. Post-Cold War international responses to
elites fearing such challenges are likely to define their
interests and security in communal terms, e.g., by de- genocidal situations have become more immediate and
forceful, but it seems that Bosnian Serb nationalists
signing policies of racial exclusion, like the Afrikaner and Rwandan militants, among others, concluded early
elite in South Africa, or advocating exclusive nation-
in the 1990s that mass killings were not likely to have
alism, like Serbs and Croats in the Yugoslav successor
states. The narrower the ethnic base of a regime, the costly international repercussions.
It also is evident that both during and after the Cold
greater the risks of conflict that escalates to genocidal War, major powers and the UN system were selec-
levels.
tive about the humanitarian crises in which to engage.
A new indicator of elite ethnicity was designed to test
There was very little international concern with geno-
the latter argument. This variable was coded based on
cides in Uganda, Rwanda, and Burundi in the 1970s
information about interethnic disputes over access to
and 1980s-countries of low economic status in which
political power. Elite ethnicity is politically salient if the the security interests of the major powers were not
ethnic or religious identity of presidents, prime minis-
threatened. At the other end of the scale, many inter-
ters, and other high officials is a recurring issue of polit- national actors were concerned about crises in Tibet
ical contention or conflict. If elite communal identity is
and Kashmir but lacked means of effective engage-
politically salient, a further distinction is made accord- ment. In the middle were Cold War-linked conflicts
ing to whether the political leadership in a given period with genocidal consequences-in Vietnam, Cambodia,
represents the largest communal group in a country or
a smaller one, using these summary codes:12 0 = elite

ethnicity categories were developed and applied by the author, using

l1 A recent discussion of the human rights consequences of cultural information in the Minorities at Risk chronologies, political hand-
diversity and analysis of alternative indicators is Walker and Poe books, and country studies. The coding was later replicated by an
2002. independent team of coders and discrepancies were resolved by ref-
12 Note that religiously defined communal groups are included in erence to additional sources. The indicators are included in the State
the scope of "ethnicity" (for example, Alawites in Syria). The elite Failure Task Force's data dictionary and data set (see footnote 7).

64
This content downloaded from 146.189.194.69 on Thu, 07 Jan 2016 05:38:52 UTC
All use subject to JSTOR Terms and Conditions
American Political Science Review Vol. 97, No. 1

Afghanistan, Angola, and Central America-in which then used to analyze data on conditions in "problem"
international rivalries trumped humanitarian concerns. countries shortly before the onset of state failure with
One underlying principle encompasses the specifics conditions in the controls.14The results are expressed as
mentioned above: The greater the degree to which a regression coefficients and as odds ratios that approxi-
country is interdependent with others, the less likely its mate the relative risks associated with each factor.15
leaders are to attempt geno-/politicides. The converse The task here is to distinguish countries where state
is that leaders of isolated states are more likely to calcu- failures led to genocides from those where they did
late that they can eliminate unwanted groups without not. The case-control method was adapted to the es-
international repercussions. Interdependency has two timation of a structural model of geno-/politicide in
dimensions, economic and political. The State Failure this manner. First, the universe of analysis consists of
studies have consistently shown that countries with a all countries already in state failure. The dependent
high degree of trade openness-indexed by exports variable represents the conditional probability that a
plus imports as a percentage of the GDP-have been genocide or politicide will begin one year later in a
less likely to experience state failures. The relationship country already experiencing failure. This avoids the
holds when controlling for population size and density problem of comparing the risks of genocide in Rwanda
and for productivity indicators. It also has the same and Sudan with, say, the negligible risks in France and
effect at the global and regional level. Moreover, trade Canada. Instead, the objective is to examine countries
openness is weakly correlated with other economic and experiencing episodes of internal wars and regime col-
trade variables. The interpretation is that trade open- lapse and determine why geno-/politicide occurred dur-
ness serves as a highly sensitive indicator of state and ing such events in Rwanda and Sudan but not, say, in
elite willingness to maintain the rules of law and fair Liberia or Nigeria. Second, the model is estimated us-
practices in the economic sphere. In the political sphere ing as cases all geno-/politicides since 1955, including
a high degree of trade openness implies that a country multiple episodes that occurred in the same country.
has more resources for averting and managing political Two of the 37 episodes were not sufficiently separated
crises (Goldstone et al. 2002). in time to be considered distinct incidents for analytic
An indicator of political interdependence is a purposes (see footnote 4).
country's memberships in regional and intercontinen-
tal organizations. Countries with greater-than-average
Results: The Final Structural Model
memberships in such organizations should be subject
to greater influence, and get more political support, The general procedure was to estimate a best-fit model
when facing internal challenge, and their regimes that included a limited set of theoretically important
should be less likely to resort to geno/politicide. Anal- variables, then to seek to improve it by testing the
ysis of the preconditions of ethnic war-a common effects of adding other variables and alternative indi-
precursor of geno-/politicide-shows that countries cators. The six-variable model summarized in Table 2
with below-average numbers of regional memberships is the culmination of a long process of model estima-
are three times more likely to have ethnic wars than tion and indicator validation. All six variables have
countries with above-average numbers of memberships significant effects at the .10 probability level; three have
(Goldstone et al. 2002).13 significant effects at the .5 level.

ESTIMATINGA STRUCTURAL MODEL 14 In

logistic regression binary indicators are used. In practice the ana-
OF GENOCIDE AND POLITICIDES lyses reported here use a mixed strategy in which most indicators-
including state failures-are binary, but others-magnitudes of
The empirical work reported here was undertaken in upheaval, trade openness, and shared memberships-are analyzed
the context of the State Failure project, which uses a as continuous variables.
case-control research design common in epidemiolog- 15 There are two widely used alternatives to the case-control method
ical research but rare in empirical social science (an for the analysis of cross-sectional time-series data on conflict. One
is simple logit analysis of pooled country-year data, both war and
exception is King and Zeng 2001b). The basic proce- peace (a recent application to civil war data is Sambanis 2001). This
dure is to match problem cases-people (or countries) approach is widely criticized because cross-national time-series con-
affected by a disorder-to a set of controls that do flict data do not meet major requirements of the method, including
not have the disorder (see Breslow and Day 1980 and nonheterogeneity and temporal independence of the observations.
It also is not well suited to analyzing rare events such as the out-
Schlesselman 1982). The State Failure project's re- break of armed conflict or, even more rare, geno-/politicide. A better
searchers selected a set of controls by matching each alternative has been proposed by Beck, Katz, and Tucker (1998),
problem case, in the year it began, with three coun- using logistic regression with dummies to achieve the equivalent of
tries that did not experience failures that year or in the event-history analysis, focusing only on intervals of peace and the
risks of transition to armed conflict. Goldstone, in a prepublication
preceding or ensuing several years. In effect, cases are paper (2002), applies both the Beck, Katz, and Tucker and the case-
selected on the dependent variable: Those experienc- control method to hypothetical and real-world data, concluding that
ing failure are matched with otherwise similar cases the latter appears to be more robust in dealing with unit and tempo-
that did not experience failure. Logistic regression is ral heterogeneity and unknown lags-which characterize the State
Failure research design and data. King and Zeng (2001a) report a
13 The reanalysis of preliminary State Failure data that focuses on problems
Regional (CIOD) and Intercontinental (CIOC) indicators of of drawing inferences from regression coefficients generated in case-
shared memberships were normalized to annual medians to eliminate control analyses. Subsequent State Failure studies, including this one,
time-trend effects. base their assessment of risks of failures on odds ratios.

65
This content downloaded from 146.189.194.69 on Thu, 07 Jan 2016 05:38:52 UTC
All use subject to JSTOR Terms and Conditions
No Lessons Learned from the Holocaust? February2003
February

TABLE 2. Final Genocide Model Results

Probabilityof
genocide given
Countriesat Countriesat Odds a single
Variable GreaterRisk Lesser Risk Ratio Coefficient p Value riskfactor
Politicalupheaval
excludingprior
genocides Higher Lower 1.70 .048 .05 .047
Priorgenocide Priorpost-1955 No prior
genocide genocide 3.39 1.220 .09 .090
Ideologicalorientation Exclusionary No exclusionary
of rulingelite ideology ideology 2.55 .937 .07 .069
Regime type Autocracies Partialor full
democracies 3.50 1.223 .03 .090
Ethniccharacterof Represents an Represents most or
rulingelite ethnic minority all groups 2.56 .939 .09 .069
Tradeopenness Lower Higher 2.58 -1.242 <.01 .070
ModelSummaryStatistics
c .83
Numberof problems 35
Numberof controls 91
Threshold .25 Settingthe thresholdat .25 ratherthan .26 increases by one the
numberof correctlyclassified genocides
% of genocides correctly 74%
classified (26 cases) Misclassifiedgenocides: Afghanistan1978, El Salvador1980, Chile
1973, Uganda 1972, Iraq1963, Yugoslavia,1992, Sudan 1956,
Philippines1972, Sri Lanka1989
% of nongenocides correctly 73%
classified (66 cases) Highest-risknongenocides:Pakistan1983, Brazil1961, Algeria
1991, China1988, Mozambique1976, Bangladesh 1974,
Indonesia1997
Note:Political upheavalandtradeopennesscoefficients arecalculatedusinginterval dataon thefullrangeof scores;othervariables
are dichotomous. Theodds ratioforupheavalis basedon the oddsassociatedwiththe seventy-fifth percentileof upheavalscores
relativeto thetwenty-fifth
percentile. Theoddsratiofortradeopennessis basedontheoddsassociatedwiththetwenty-fifth percentile
of opennessscoresrelativeto the seventy-fifth percentile. of geno-/politicide
Theprobability fora countrywithno riskfactorsis .028.
Probabilitiesassumethata country instatefailure.
is currently Thecstatisticis usedtocomparemodelsincase-controlanalysiswithout
limitingthe comparisonto one particular choiceof modelthresholds(GreenandSwets1966;Harrell et al. 1984).Itis the fraction
of
concordant pairsinthedata,i.e.,thefraction of allpossibleproblemandcontrolpairsforwhichthemodelassignsa higherscoreto the
problem thanthecontrol.

Political Upheaval. Consistent with the theoretical geno-/politicides. The effects of magnitude of po-
argument, the greater the magnitude of previous in- litical upheaval were weaker than those of prior
ternal wars and regime crises, summed over the pre- genocide-it appears that habituation to genocide adds
ceding 15 years, the more likely that a new state fail- more to the risks of future genocide than the mag-
ure will lead to geno-/politicide. When the magnitude nitude of internal war and adverse regime change
of past upheaval was divided between high and low, per se.
in high-magnitude cases the risks of geno-/politicides
were nearly two times greater. Similar results were Elite Ideology and Regime Type. Theoretical argu-
obtained using other indicators of upheaval, e.g., the ments about the importance of elite ideologies and
magnitudes of armed conflict within states.16 regime type are supported. Countries in which the rul-
ing elite adhered to an exclusionary ideology were two
Prior Genocides. Arguments about the recurrence of and a half times as likely to have state failures lead-
geno-/politicide also are supported. The risks of new ing to geno-/politicide as those with no such ideology.
episodes were more than three times greater when Failures in states with autocratic regimes were three
state failures occurred in countries that had prior and a half times more likely to lead to geno-/politicides
than failures in democratic regimes. These events, if
they occur in the context of other risk factors identified
16 A number of alternative indicators of in the model, substantially raise the risks of massive
political upheaval were
tested. The best results are obtained using the full range of values human rights violations.
on an indicator that sums annual magnitudes of state failure for the
previous 15 years. The range of the indicator is 0 to 46 (mean, 8.0; Ethnic and Religious Cleavages. Numerous indica-
median, 4.5). To facilitate interpretation Table 2 reports an odds ratio
based on the odds of geno-/politicide associated with the seventy-fifth tors of ethnic and religious cleavages were evaluated
percentile of upheaval scores relative to the twenty-fifth percentile. but only one was significant in the final model. The

66
This content downloaded from 146.189.194.69 on Thu, 07 Jan 2016 05:38:52 UTC
All use subject to JSTOR Terms and Conditions
American Political Science Review Vol. 97, No. 1

risks of geno-/politicide were two and a half times more Burundi fit this pattern, i.e., they are challenged by
likely in countries where the political elite was based rebels (are in failure) and have all risk factors except an
mainly or entirely on an ethnic minority. exclusionary ideology. An example in the lower range
of risks is a hypothetical failed state with high political
International Interdependences. Countrieswith low
trade openness had two and a half times greater odds upheaval and low trade openness but no other risk fac-
tors: The conditional probability of geno-/politicide in
of having state failures culminate in geno-/politicide.17 such a state is .11 (confidence interval, .04 to .27). Both
High trade openness (and the underlying economic and Colombia and India have this pattern. If such a state is
political conditions it taps) not only minimizes the risks also autocratic, the conditional probability increases to
of state failure in general, as shown in other State Fail- .30 (confidence interval, .16 to .50).
ure analyses, but reduces substantially the odds that Efforts to validate the model in Table 2 included
failures, if they do occur, will lead to geno-/politicides.
The results support arguments about the importance of refining the list of problem and control cases (not re-
a country's international economic linkages in inhibit- ported here), assessing the effects of interdependence
of cases of geno-/politicides, probing the robustness of
ing gross human rights violations. Results of analyses theoretical findings with different data and alternative
using indicators of political interdependence are pre- indicators, and testing the effects of additional risk fac-
sented below. tors. The final model that emerges from this process
The overall accuracy of the model is shown in the
displays less classificatory accuracy than some prior
summary statistics and notes. The c statistic is used to models but engenders greater confidence in terms of
summarize and compare the discriminatory power of theoretical plausibility, accuracy, and reliability as an
different models in case-control analysis. It is the pro- instrument for generating risk assessments.
portion of all possible problem and control pairs for The purpose of posting the State Failure data set and
which the model generates a higher score for the prob-
lem than the control. Coefficients from case-control dictionary on a web site (see footnote 7) is to enable
other researchers to apply other techniques and test
regression analysis are used to generate model scores. alternative models-a number of which have already
Typically a threshold is chosen for the range of model been done by the Task Force. Following are the results
scores that equalizes the proportions of correctly clas- of a few of the alternative case-control analyses done
sified problems and controls. The model correctly clas- to assess the validity of the geno-/politicide model.
sifies 74% of all cases as genocides or nongenocides.
The right-hand column in Table 2 shows the con-
ditional effects of each variable on the probability of
Probing Effects of Interdependence
geno-/politicide in a country with an internal war or
regime crisis. Note that, since the analyses include Interdependence among independent variables and
all cases of state failure and all instances of geno-/ among cases of geno-/politicide are potential threats
politicide, issues of sampling error do not arise. The to the stability of the model. Bivariate tests of rela-
probability of genocide for a country in failure with no tionships among the independent variables showed few
risk factors is .028. If the country is an autocracy but significant associations. The exceptions were that ex-
has no other risk factors, the probability is increased clusionary ideologies were twice as common among
by .090. If the country has a minority elite but no other autocracies as among democracies, and the magnitude
risk factors, the probability is increased by .069. of past upheaval was significantly higher in autocracies
The incremental effects of each risk factor are rel- and in regimes with exclusionary ideologies. However,
atively small; their cumulative effect is large. Analysis these exceptions are not strong enough to affect the
of various combinations of risk factors shows that, if stability of the model.
all risk factors are present in a failed state, the con- The fact that multiple genocides occurred in
ditional probability of geno-/politicide is .90, with a ten countries raises questions about the statistical inde-
95% confidence interval of .66 to .98. The only such pendence of cases. This was tested by reestimating the
country with all six factors in 2001 is Iraq, as shown model with a reduced set of cases: Only the first geno-
in Table 5. A hypothetical country with the following /politicide in a country was included; subsequent cases
combination of four risk factors-a high magnitude of were excluded. This had virtually no effect on odds ra-
past upheaval, a minority elite, low trade openness, and tios, the c statistic, or the classification accuracy. The
autocracy-has a conditional geno-/politicide probabil- conclusion from these and other analyses not reported
ity of .52 (confidence interval, .27 to .77). Sierra Leone is here is that the model is not significantly affected by
a contemporary example. If such a failed state also had interdependencies.
a past genocide, the probability increases to .79 (confi-
dence interval, .43 to .95). Contemporary Rwanda and
Probing Societal Cleavage and Elite
Ethnicity Effects
17
Trade openness is used as a continuous variable for calculating The study is particularly concerned with the effects
coefficients and estimating model scores. The range of the indicator is of alternative patterns of ethnic diversity and ethnic
7 to 137 (mean, 49; median, 43). For purposes of reporting odds ratios
in Table 2 for this variable the odds of geno-/politicide associated with discrimination on the risks of geno-/politicide. For ex-
the twenty-fifth percentile of openness scores are shown relative to ample, the (COW) Cultural data set was used to con-
the seventy-fifth percentile. struct dichotomous indicators of low, medium, and

67
This content downloaded from 146.189.194.69 on Thu, 07 Jan 2016 05:38:52 UTC
All use subject to JSTOR Terms and Conditions
No Lessons Learned from the Holocaust? February
February2003

TABLE 3. Probing the Structural Model: Testing Effects of Alternative Elite and Societal
Cleavage Indicators
ModelwithSalient Modelwith Modelwith Ethnic
EliteEthnicity MinorityElite Discrimination
Variableand Predicted
Directionof Effect Odds Ratio p Value Odds Ratio p Value Odds Ratio p Value
Politicalupheavalexcluding
priorgenocides (+) 1.65 .05 1.70 .05 1.66 .08
Priorgenocide (+) 3.30 .11 3.39 .09 2.71 .17
Exclusionaryelite ideology (+) 2.26 .11 2.55 .07 2.35 .10
Autocracy(+) 3.38 .03 3.40 .03 3.66 .03
Eliteethnicityis salient (+) (1.48) (.53)
Eliteis ethnoculturalminority(+) 2.56 .09 2.59 .09
Activeethnicdiscrimination (2.08) (.24)
Tradeopenness (-) 2.43 <.01 2.56 <.01 2.53 <.01
ModelSummaryStatistics
c .83 .83 .84
Numberof problems 35 35 35
Numberof controls 91 91 91
Threshold .24 .25 .26
% of genocides correctlyclassified 74% (26 cases) 74% (26 cases) 77% (27 cases)
% of nongenocides correctlyclassified 76% (69 cases) 73% (66 cases) 75% (68 cases)
Note: See Table2 for explanatorynotes. Onlyodds ratiosand p values are shown here to simplifycomparisonacross alternative
models.

high ethnic diversity. The low-diversity indicator tests organizations (NGOs), and reactive or preventive
whether small ethnic groups are likely to be targeted; military action (see Heidenreich 2001, Kuper 1985,
the medium-diversity indicator provides a test of the and Riemer 2000). To test whether international
thesis that the presence of several relatively large ethnic political linkages are more important than economic
groups is associated with relatively intense interethnic interdependency in checking the escalation of internal
contention and genocide. Bivariate analyses showed wars and adverse regime changes into geno-/politicides,
that neither pattern of ethnic diversity had any effect on two indicators of countries' shared membership in
the likelihood of geno-/politicide. This supports Krain's international organizations were constructed. The
(1997) empirical finding that ethnic fractionalization assumption is that the greater a country's international
was uncorrelated with the onset of geno-/politicide. It political connections, the more likely it is to accept help
is also consistent with Walker and Poe's (2002) cross- in managing conflict and the more susceptible it is to
national analysis showing that cultural heterogeneity pressures to minimize human rights violations. Indica-
was not related to political repression. tors of country memberships in regional and intercon-
But two other indicators of ethnic cleavage did cor- tinental (i.e., global) organizations were substituted in
relate significantly (in bivariate analysis) with geno-/ the structuralmodel for trade openness, with the results
politicide: active ethnic discrimination and elite eth- shown in Table 4. Neither is statistically significant,
nicity. To assess their effects better, three alternative and neither adds to the classificatory accuracy of the
models were estimated as shown in Table 3. The first model, but both reduce the impact of the minority elite
two models differ in that the first uses the "elite eth- indicator. This suggests that minority elites are less
nicity is salient" indicator, while the second uses the likely to have dense networks of political ties with
"elite represents an ethnic minority" indicator. "Elite other countries, and thus are more free to commit
represents an ethnic minority" is significant, whereas human rights violations, but the direction of causality
the "salience" indicator is not. Ethnic discrimination is is uncertain.
added to the model in the last column in Table 3. It has This does not mean that political linkages are irrele-
weak effects in the predicted direction (p = .24). Thus vant to the analysis and prevention of geno-/politicide.
the final model includes only the "elite ethnic minor- The critical question, in my view, is whether states and
ity" indicator. The theoretical implication is important: international organizations do in fact engage in pre-
Ethnic heterogeneity is likely to lead to geno-/politicide ventive actions in the early stages. What counts is the
only if an ethnic minority dominates the elite. political will to engage-and as Powers (2002) shows
in her recent study, major international actors have re-
Probing International Interdependence peatedly chosen not to do so.
and Quality-of-Life Effects Finally, the indicator of infant mortality was
evaluated in the model because other analyses
Genocide scholars almost invariably advocate interna- consistently show that high infant mortality rates
tional action including preventive diplomacy by the UN contribute to the risks of state failure (Goldstone
and major powers, engagement by nongovernmental et al. 2002). The results in the last two columns in

68
This content downloaded from 146.189.194.69 on Thu, 07 Jan 2016 05:38:52 UTC
All use subject to JSTOR Terms and Conditions
American Political Science Review Vol. 97, No. 1

TABLE 4. Revising the Structural Model: Testing Effects of International Political

Interdependencies and Quality of Life
RegionallOs Intercontinental
lOs InfantMortality
Replaces Trade Replaces Trade Addedto Modelwith
Openness Openness TradeOpenness
Variable and Predicted
Directionof Effect Odds Ratio p Value Odds Ratio p Value Odds Ratio p Value
Politicalupheavalexcludingprior
genocides (+) 1.53 .10 1.60 .06 1.65 .08
Priorgenocide (+) 3.93 .05 3.78 .05 3.50 .08
Exclusionaryelite ideology (+) 2.47 .07 2.47 .07 2.47 .09
Autocracy(+) 2.93 .05 2.85 .06 3.45 .04
Eliteis ethnoculturalminority(+) (1.66) (.33) (1.73) (.30) 2.64 .10
Tradeopenness (-) 2.60 <.01
Intercontinental10 memberships(+) (1.00) (.99)
Regional 10 memberships(+) (1.38) (.33)
Infantmortality(-) (.94) (.89)
ModelSummaryStatistics
c .79 .80 .83
Numberof problems 35 35 34
Numberof controls 91 91 91
Threshold .25 .23 .26
% of genocides correctlyclassified 71% (25 cases) 71% (25 cases) 74% (25 cases)
% of nongenocides correctlyclassified 71% (65 cases) 71% (65 cases) 72% (66 cases)
Note: See Table2 for explanatorynotes. Only odds ratiosand p values are shown here to simplifycomparisonacross alternative
models. 10, internationalorganization.

Table 4 show that, unlike trade openness, infant mor- tively ended Philippine democracy by declaring martial
tality (and the basket of quality-of-life indicators law.
it represents) has no independent effects on the Thus, five of the nine misclassified cases can be ac-
odds of geno-/politicide once a state failure has oc- counted for by variables within the model. An example
curred. The conclusion is that low development is a of an exception is Iraq in 1963, when the military first
generic risk factor for civil conflict and regime instabil- resorted to indiscriminant violence to suppress a rebel-
ity, whereas the effects of economic interdependence lion by separatist Kurds;it may be questioned whether
are both global and specific to the risks of geno-/ the case meets the key criterion of intent to destroy
politicide. the group. If five cases that are misclassified because
of temporal inconsistencies in the data and research
False Negatives, False Positives design are added to the 26 correctly classified geno-/
politicides, the classification accuracy increases to a
The nine misclassified cases identified at the bottom very substantial 89%.
of Table 2 ("misclassified genocides") are diagnostic The false positives, in contrast, are the control cases
of problems with data error and model specifica- of state failure incorrectly classified as impending geno-
tion. Sudan 1956 is misclassified, probably because the cides. Case-by-case inspection suggests that several
genocide is dated from the beginning of the Southern could easily have escalated into politicides. Algeria af-
rebellion, whereas mass killings began an indetermi- ter the onset of militant Islamic insurgency in 1991 is
nant number of years later. If the onset of geno- a case in point. Algeria's Islamic militants, a rival au-
cidal policies were dated from the mid-1960s, when thority in a civil war situation, carried out a campaign
the regime was no longer democratic and the mag- of murder against civilians but seem not to have tar-
nitude of upheaval was high, it would be correctly geted any specific group; therefore the campaign does
classified. There were similar lags before the on- not qualify as a politicide. Similarly, in Mozambique in
set of geno-/politicides in Afghanistan (1978) and 1976, widespread killings were carried out by Renamo
El Salvador (1980). Two other cases are misclassified rebels, but they did not target a definable communal
because of the lag structure in the data used to es- group. The most intriguing of the recent false posi-
timate the model: Chile 1973 (targeting the left) and tives is Indonesia. The failure was the dissolution of
the Philippines 1972 (targeting rebellious Moros). Both the Suharto regime in 1997 and the model shows a high
regimes are classified as democracies because all model risk of impending genocide. In fact, it took international
variables are measured one year prior to the onset of intervention in East Timor to prevent a resumption of
geno-/politicide. In fact, both geno-/politicides were genocidal killings following the independence referen-
carried out by authoritarian leaders after they sus- dum of 1999, and the risks of a future episode in Aceh
pended democratic rule: General Pinochet overthrew province remain high (see Genocide Prevention Center
the democratic Allende government, and Marcos effec- 2001).

69
This content downloaded from 146.189.194.69 on Thu, 07 Jan 2016 05:38:52 UTC
All use subject to JSTOR Terms and Conditions
No Lessons Learned from the Holocaust? February2003
February

CONCLUSION: USING THE MODEL TO active groups is likely to prolong internal wars and
ASSESS RISKS OF FUTURE GENOCIDES may increase the risks of geno-/politicide. Also, empty
AND POLITICIDES threats from the international community against per-
petrators may signal to elites that they can get away
All episodes of genocide and political mass murder of with mass murder without international repercussions.
the last half-century have been carried out by elites Another promising avenue for research is the "acceler-
or rival authorities in the context of internal war and ator" approach in which event data analysis is applied
regime instability. The motive common to such elites to pregenocidal situations to identify patterns and se-
is the destruction "in whole or part" of collectivities
quences of political events that signal the onset of mass
that challenge their claim to authority or stand in the
killings (see Harff 2001 and Harff and Gurr 1998).
way of an ideology-driven desire to create a society
purified of undesirable classes or communal groups.
The structural model tested here identifies six causal Policy Implications
factors that jointly differentiate with 74% accuracy the
35 serious civil conflicts since 1955 that led to episodes The model provides a framework for assessing and
of genocide and politicide from 91 others that did not comparing the vulnerability of countries with state fail-
have genocidal consequences. The risk factors include ures to genocide and politicide. When the model is ap-
the extent of political upheaval and the occurrence of plied to current information, it provides the basis for
a global "watch list" that identifies countries in which
prior geno-/politicides. The probability of mass murder the conditions for a future episode are present.
is highest under autocratic regimes and is most likely to
be set in motion by elites who advocate an exclusionary Table 5 provides an illustration. It lists 25 countries
with armed conflicts circa 2001, based on an ongoing
ideology, or represent an ethnic minority, or both. Inter-
national economic interdependencies sharply reduce survey by Marshall (see Gurr, Marshall, and Khosla
the chances that internal war and regime instability will 2001). Each country is categorized on the six risk fac-
have genocidal consequences. tors using data for the year 2000 compiled by the State
Failure project. Possible victim groups are identified
based on country-specific information on civil conflicts,
Implications for Theory minority groups, and political opposition movements.
The countries are ranked according to their numbers
Some theoretical arguments about the causes of geno- of positive risk factors.
cide are called into question by the results. First, indi- Eleven countries currently are high on four or more
cators of ethnic and religious cleavages had ambiguous of the risk factors for genocide and politicide. Iraq is
effects in the final model. Active discrimination against the only one in which all six risk factors are present;
ethnic minorities is a significant causal factor leading Afghanistan also had all six factors prior to the over-
to ethnic war, consistent with theories of ethnic con- throw of the Taliban regime but still faces considerable
flict (Gurr 2000; Horowitz 1985), but once ethnic and risks depending on the outcome of efforts to recon-
other civil wars have begun, discrimination does not struct a nonideological coalition government. The three
help explain which of them are likely to lead to geno-/ countries with five factors are Burma, Burundi, and
politicide. Second, levels of economic development, in- Rwanda. Another five African countries have four fac-
dexed here by infant mortality, make no difference in tors, as does Algeria. China, responsible for three previ-
the likelihood of geno-/politicide once internal wars ous episodes, has three risk factors, along with Liberia
and adverse regime changes have begun. Low eco- and Pakistan. For this approach to be useful for pol-
nomic development and societal cleavages may predis- icy analysis, tabulation of risk factors must be comple-
pose societies to intense conflict but it is characteristics mented by assessment of the political circumstances
of the political system and elites, and a country's inter- in which they might be activated, as suggested in the
national linkages, that determine whether those con- following sketches of Burma and China.
flicts culminate in geno-/politicide. It was also found,
contrary to expectations, that economic interdepen- Burma. The SLORC (State Law and Order Restora-
dence is more important than international political tion Council), the military council that controls Burma,
linkages. The reason, I suggest, is that the international has relied mainly on repression to control its domestic
will to act is more important than political linkages in opponents, including communal separatists such as the
preventing escalation to geno-/politicide. Karen and Shan and the urban-based democratic op-
The findings are all the more significant in light of position. The government also targeted the Muslims
the fact that it was not possible to test effects of some of the northwest Arakan region for destruction or ex-
other causal factors postulated by genocide scholars for pulsion in 1978 and, again, in the early 1990s, precipi-
which comparative historical data are currently lack- tating large-scale refugee flows. The potential that any
ing. In future structural analyses it would be desir- of these conflicts might escalate into geno-/politicide
able to assess the presence and expansion of private is moderated by several factors. First, the SLORC has
militias, paramilitary units, and state security agencies sought negotiated settlements with most regional sepa-
that operate with few restraints. Such entities often are ratists, indicative of a shift away from exclusive reliance
implicated in the targeting of civilians. External sup- on repression. Second, it is seeking an accommoda-
port for a targeted group also has complex effects that tion with the democratic opposition, partly in response
merit closer analysis. Material support for politically to international pressures. Finally, the country is being

70
This content downloaded from 146.189.194.69 on Thu, 07 Jan 2016 05:38:52 UTC
All use subject to JSTOR Terms and Conditions
TABLE 5. Risks of Genocide and Politicide in Countries with Armed Conflicts in 2001
Risk Factor
Countries (No. of Upheaval Exclusionary Trade
Risk Factors) a PriorGeno-/Politicides since 1986b MinorityElite Ideology Typeof Regime Opennessc
Iraq(6 of 6) Yes: 1961-75, High Yes: Sunni Arabs Yes: Secular Autocracy Verylow Kur
1988-91 dominate nationalist
Afghanistan2000 Yes: 1978-89 Veryhigh Yes: Pushtuns Yes: Islamist Autocracy Verylow Haz
(6 of 6) dominate
Afghanistan2002 Yes: 1978-89 Veryhigh No: Coalitionin No No effective Verylow Sup
(4 of 6) formation regime
Burma(5 of 6) Yes: 1978 High No: Burmanmajority Yes: Nationalist Autocracy Verylow Dem
dominates A
Burundi(5 of 6) Yes: 1965-73, 1993, 1998 Veryhigh Yes: Tuisis dominate No Autocracy Low Hut
Rwanda(5 of 6) Yes: 1963-64, 1994 High Yes: Tutsidominate No Autocracy Low Sup
Congo-Kinshasa Yes: 1964-65, 1977 Veryhigh Yes: Narrowcoalition No No effective Medium Hut
(4 of 6) of Kabilasupporters regime of
Somalia(4 of 6) Yes: 1988-91 Veryhigh No: Clan rivalries No No effective Verylow Isaa
regime
SierraLeone No Veryhigh Yes: Mende No No effective Low Sup
(4 of 6) dominated regime U
Ethiopia(4 of 6) Yes: 1976-97 High Yes: Tigreans dominated No Autocracy Medium Sup
Uganda(4 of 6) Yes: 1972-79, 1980-86 High No No Autocracy Low Sup
Algeria(4 of 6) Yes: 1962 Veryhigh No Yes: Secular Autocracy Medium Isla
nationalist
Liberia(3 of 6) No High No No Autocracy Low Kra
Ta
Pakistan(3 of 6) Yes: 1971, 1973-77 Medium No: Punjabimajority No Autocracy Low Sind
dominates
China(3 of 6) Yes: 1950-51, 1959, Medium No Yes: Marxist Autocracy Medium Uigh
1956-75
Sri Lanka(2 of 6) Yes: 1989-90 High No: Sinhalese majority No Partialdemocracy High Sup
dominates
Philippines(2 of 6) Yes: 1972-76 Veryhigh No No Democracy High Sup
Colombia(2 of 6) No Veryhigh No No Partialdemocracy Low Pea
Turkey(2 of 6) No High No Yes: Secular Partialdemocracy Medium Kur
nationalist
India(2 of 6) No High No No Democracy Low Mus
Israel(2 of 6) No Veryhigh No Yes: Ethnic Democracy High Pale
nationalism
Indonesia(1 of 6) Yes: 1965-66, 1975-92 Medium No: Javanese dominate No Partialdemocracy Medium Ace
Russia (1 of 6) Yes: 1943-45, 1947-50 Low No No Partialdemocracy Medium Che
Nigeria(0 of 6) No Low No: Muslimmajority No Partialdemocracy High Chr
dominates
Nepal (0 of 6) No Medium No No Partialdemocracy Medium Mao
Macedonia(0 of 6) No None No: Macedonianmajority No Partialdemocracy High Ethn
rdomintQOe
a Countriesare listed
accordingto theirnumberof riskfactors,as identifiedin the finalmodel in Table2. Priorgenocides are fromHarff1992; otherd
Allindicatorsuse 2000 informationexcept that trade openness is for 1999. Boldfaceitalicentries are high-riskconditions.The table includes most co
(fromGurr,Marshall,and Khosla2001) butexcludes Angolaand Sudan because they had ongoing geno-/politicidesin 2001. Possible victimgroupsa
informationfromcase files on countries,minoritygroups,and politicaloppositionmovements.
bCategoriesused forupheavalscores: low, 1-9; medium,10-20; high,21-34; very high,35-60.
CCategoriesused fortradeopenness scores: very low,2 or less; low,21-40; medium,41-70; high,71-100; very high, >100.

This content downloaded from 146.189.194.69 on Thu, 07 Jan 2016 05:38:52 UTC
All use subject to JSTOR Terms and Conditions
No Lessons
No Lessons Learned from the
Learned from the Holocaust?
Holocaust? February2003
February 2003

opened up to foreign investment in ways that will con- Binary Dependent Variable."American Journal of Political Science
tinue to reduce its rulers' future options about how to 42 (4): 1260-88.
deal with opponents. Breslow, N. E., and N. E. Day. 1980. Statistical Methods in Cancer
Research. Volume I: The Analysis of Case-Control Studies. Lyon:
IARC Scientific Publications No. 32.
China. Three geno-/politicides have been carried out Chalk, Frank, and Kurt Jonassohn. 1990. The History and Sociology
during Communist rule in China, one after the Com- of Genocide: Analyses and Case Studies. New Haven, CT: Yale
munists took power in 1950-51, the second in Tibet University Press.
in 1959, and the last during the Cultural Revolution Davenport, Christian. 1999. "Human Rights and the Democratic
Proposition." Journal of Conflict Resolution 43 (1): 92-116.
from 1966 to 1975. Some risk factors remain high but Des Forges, Allison. 1999. Leave None to the Tell the Story: Genocide
appear to be declining. Beijing's rulers are more prag- in Rwanda. New York: Human Rights Watch.
matic in doctrine and practice than their predecessors. Esty, Daniel C., Jack A. Goldstone, Ted Robert Gurr, Barbara Harff,
Marc Levy, Geoffrey D. Dabelko, Pamela T. Surko, and Alan N.
China is more engaged economically with the rest of the
Unger. 1999. "State Failure Task Force Report: Phase II Findings."
world, with the likely long-term result of constraining Environmental Change & Security Project Report 5 (Summer):
domestic policies that offend trading partners and in- 49-72.
vestors. Nonetheless, the regime has responded harshly Fein, Helen. 1979. Accountingfor Genocide: National Responses and
to resistance by Tibetans and by Muslim Uighers in Jewish Victimization during the Holocaust. New York: Free Press.
Fein, Helen. 1984. "Scenarios of Genocide: Models of Genocide and
Xinjiang province and to imagined security threats Critical Responses." In Toward the Understanding and Preven-
from Christians and the Falun Gong movement. Un- tion of Genocide: Proceedings of the International Conference on
less the Chinese government becomes more willing to the Holocaust and Genocide, ed. Israel W. Charny. Boulder, CO:
accommodate national minorities and unauthorized re- Westview Press, 3-31.
Fein, Helen. 1993a. "Accounting for Genocide After 1945: Theories
ligious sects, the risk remains that repression may es- and Some Findings." InternationalJournal on Group.Rights 1: 79-
calate into policies aimed at eliminating the offending 106.
groups. Fein, Helen 1993b. Genocide:A Sociological Perspective.London and
These two sketches help make another point about Newbury Park, CA: Sage.
the policy implications of the structural model. Some Genocide Prevention Center. 2001. Aceh: Indonesia Report. Impend-
factors are historically inescapable, including the oc- ing Clash May Result in Crimes against Humanity. Alexandria, VA:
Improve the World International.
currence of prior genocides, but most are susceptible to Goldstone, Jack A. 1991. Revolution and Rebellion in the Early Mod-
external influence. For example, promoting the obser- ern World. Berkeley and Los Angeles: University of California
vance of minimal human rights standards and the prac- Press.
tice of inclusiveness should continue to be on the pol- Goldstone, Jack A. 2002. "The Case for Case Control: Substantive
Issues in the Study of Rare and Calamitous Events." University
icy agenda of governments and organizations that care of California, Davis. Unpublished paper prepared for the State
what happens in countries such as Burma and Burundi. Failure Task Force.
Attempts to force democratization are problematic be- Goldstone, Jack A., Ted Robert Gurr, Barbara Harff, Marc
cause such attempts in poor, heterogeneous countries Levy, Monty G. Marshall, Robert Bates, David Epstein, Colin
Kahl, Mark Woodward, Pamela T Surko, and Alan N. Unger.
often fail. Nevertheless, those failures usually prompt 2002. State Failure Task Force Report: Phase III Findings.
efforts to redesign and rebuild democratic institutions. McLean, VA: Science Applications International Corporation. At
Once in place, democratic institutions-even partial http://www.cidcm.umd.edu/inscr/stfail.
ones-reduce the likelihood of armed conflict and all Green, D. M., and J. A. Swets. 1966. Signal Detection Theory and
but eliminate the risk that it will lead to geno-/politicide. Psychophysics. New York: John Wiley and Sons.
Gunaratna, Rohan. 1990. Sri Lanka: A Lost Revolution? The In-
Moreover, economic connectedness appears to have at side Story of the JVP. Kandy: Sri Lanka Institute of Fundamental
least one positive effect in poor countries: The greater Studies.
their interdependence with the global economy, the less Gurr, Ted Robert, ed. 1980. Handbook of Political Conflict: Theory
and Research. New York: Free Press.
likely that their elites will target minorities and political Gurr, Ted Robert. 1986. "Persisting Patterns of Repression and Re-
opponents for destruction. bellion: Foundations for a General Theory of Political Coercion."
The risk assessments generated using this approach In PersistentPatternsand Emergent Structuresin a Waning Century,
not only signal possible genocides, but flag the actual ed. Margaret. P Karns. New York: Praeger, 149-68.
and potential victims of human rights abuses in conflict- Gurr, Ted Robert. 2000. Peoples versus States: Minorities at Risk in
the New Century. Washington, DC: U.S. Institute of Peace Press.
ridden countries everywhere. Timely and plausible as-
Gurr, Ted Robert, Monty G. Marshall, and Deepa Khosla. 2001.
sessments of these situations should make it easier to Peace and Conflict 2001: A Global Survey of Armed Conflicts,Self-
convince policymakers of the need to engage proac- Determination Movements, and Democracy. College Park, MD:
tively in high-risk situations. Anticipatory responses Center for International Development and Conflict Management.
should save more lives at less cost than belated re- Harff, Barbara. 1987. "The Etiology of Genocide." In The Age
of Genocide, ed. Michael N. Dobkowski and Isador Wallimann.
sponses after killings have begun. Westport, CT: Greenwood Press, 41-59.
Harff, Barbara. 1992. "Recognizing Genocides and Politicides." In
Genocide Watch,ed. Helen Fein. New Haven, CT:Yale University
REFERENCES Press, 27-41.
Harff, Barbara. 2001. "Could Humanitarian Crises Have Been An-
Ball, Patrick, Paul Kubrick,and Herbert F Spirer. 1999. State Violence ticipated in Burundi, Rwanda, and Zaire? A Comparative Study of
in Guatemala, 1960-1996: A Quantitative Reflection. Washington, Anticipatory Indicators." In Journeys through Conflict:Narratives
DC: American Association for the Advancement of Science. and Lessons, ed. Hayward Alker, Ted Robert Gurr, and Kumar
Banks, Arthur S. 1971. Cross-Polity Time-Series Data. Cambridge, Rupesinghe. Lanham, MD: Rowman & Littlefield, 81-102.
MA: MIT Press. Harff, Barbara, and Ted Robert Gurr. 1998. "Systematic Early Warn-
Beck, Nathaniel J., Jonathan N. Katz, and Richard Tucker. 1998. ing of Humanitarian Emergencies." Journal of Peace Research 35:
"TakingTime Seriously: Time-Series-Cross Section Analysis with a 551-79.

72
This content downloaded from 146.189.194.69 on Thu, 07 Jan 2016 05:38:52 UTC
All use subject to JSTOR Terms and Conditions
American Political
American Science Review
Political Science Review Vol.
Vol. 97,
97, No.
No. 1

Harrell, E E., Jr., K. L. Lee, R. M. Califf, D. B. Pryor, and R. A. Rosati. Lopez, George A. 1986. "National Security Ideology as an Impetus
1984. "Regression Modeling Strategies for Improved Prognostic to State Violence and State Terror."In Government Violence and
Prediction." Statistics in Medicine 3 (2): 143-52. Repression: An Agenda for Research, ed. George A. Lopez and
Heidenreich, John G. 2001. How to Prevent Genocide: A Guide for Michael Stohl. Westport, CT: Greenwood Press, 73-95.
Policymakers, Scholars, and the Concerned Citizen. Westport, CT: Makiya, Kanan. 1992. "The Anfal: Uncovering an Iraqi Campaign to
Praeger. Exterminate the Kurds."Harper's Magazine 284, No. 1704 (May):
Hegre, Havard, Tanja Ellingsen, Scott Gates, and Nils Petter Gled- 53-61.
itsch. 2001. "Toward a Democratic Civil Peace? Democracy, Polit- Mattas, David. 1992. "Remembering the Holocaust." In Geno-
ical Change and Civil War, 1916-1992." American Political Science cide Watch, ed. Helen Fein. New Haven, CT: Yale University
Review 95 (March): 33-48. Press.
Henderson, Conway. 1991. "Conditions Affecting the Use of Political Melson, Robert. 1992. Revolution and Genocide. Chicago: University
Repression." Journal of Conflict Resolution 35: 120-42. of Chicago Press.
Horowitz, Donald L. 1985. Ethnic Groups in Conflict. Berkeley and Minority Rights Group. 1984. Uganda and Sudan-North and South.
Los Angeles: University of California Press. Report No. 66. London: Minority Rights Group.
Human Rights Watch. 1991. "Sri Lanka." In World Report 1990. Munck, Gerardo L., and Jay Verkuilen. 2002. "Conceptualizing and
New York: Human Rights Watch. Measuring Democracy: Evaluating Alternative Indices." Compar-
Jaggers, Keith, and Ted Robert Gurr. 1995. "Tracking Democracy's ative Political Studies 35: 5-34.
Third Wave with the Polity III Data." Journal of Peace Research Powers, Samantha. 2002. "A Problem from Hell": American and the
32 (November): 469-82. Age of Genocide. New York: Basic Books.
Kasozi, A. B. K. 1994. The Social Origins of Violence in Uganda, Prunier, Gerard. 1995. The Rwanda Crisis: History of a Genocide.
1964-1985. Montreal: McGill-Queen's University Press. New York: Columbia University Press.
King, Gary, and Langche Zeng. 2001a. "Improving Forecasts of State Quigley, T., trans. (1999). Guatemala: Never Again. Report of the
Failure." World Politics 53 (July): 623-58. Guatemalan Roman Catholic Church's Inter-Diocesan Recupera-
King, Gary, and Langche Zeng 2001b. "Logistic Regression in Rare tion of the Historical Memory Project (REMHI). Maryknoll, NY:
Events Data." Political Analysis 9 (Spring): 137-63. Orbis Books.
Krain, Matthew. 1997. "State-Sponsored Mass Murder: The Onset Rabushka, Alvin, and Kenneth A. Shepsle. 1972. Politics in Plu-
and Severity of Genocides and Politicides." Journal of Conflict ral Societies: A Theory of Democratic Instability. Columbus, OH:
Resolution 41: 331-60. Merrill.
Kuper, Leo. 1977. The Pity of ItAll: Polarization of Racial and Ethnic Riemer, Neil, ed. 2000. Protection Against Genocide: Mission Impos-
Relations. Minneapolis: University of Minnesota Press. sible? Westport, CT: Praeger.
Kuper, Leo. 1981. Genocide: Its Political Use in the TwentiethCentury. Rummel, Rudolph J. 1995. "Democracy, Power, Genocide, and Mass
New Haven, CT: Yale University Press. Murder."Journal of Conflict Resolution 39: 3-26.
Kuper, Leo. 1985. The Prevention of Genocide. New Haven, CT:Yale Sambanis, Nicholas. 2001. "Do Ethnic and Nonethnic Civil Wars
University Press. Have the Same Causes? A Theoretical and Empirical Inquiry
Le Blanc, Lawrence J. 1988. "The United Nations Convention and (Part I)." Journal of Conflict Resolution 45 (3): 259-82.
Political Groups: Should the United States Propose an Amend- Schlesselman, J. J. 1982. Case-Control Studies:Design, Conduct, Ana-
ment?" Yale Journal of International Law 13: 268-94. lysis. New York: Oxford University Press.
Lemkin, Raphael. 1944. Axis Rule in Occupied Europe. Washington, Schmeidl, Suzanne. 1997. "Exploring the Causes of Forced Migration:
DC: Carnegie Endowment for International Peace. A Pooled Analysis, 1971-1990." Social Science Quarterly 78: 284-
Lichbach, Mark I. 1987. "Deterrence or Escalation? The Puzzle of 308.
Aggregate Studies of Repression and Dissent." Journal of Conflict Tilly, Charles. 1978. From Mobilization to Revolution. Reading, MA:
Resolution 31: 266-97. Addison-Wesley.
Lichbach, Mark I. 1998. "Contending Theories of Contentious Totten, Samuel, and Steven Jacobs, eds. 2002. Pioneers of Geno-
Politics and the Structure-Action Problem of Social Order." An- cide Studies: Confronting Mass Death in the Century of Genocide.
nual Review of Political Science 1: 401-24. Westport, CT: Greenwood Press.
Licklider, Roy. 1995. "The Consequences of Negotiated Settlements Walker, Scott, and Steven C. Poe. 2002. "Does Cultural Diversity
in Civil Wars, 1945-1993." American Political Science Review 89: Affect Countries' Respect for Human Rights?" Human Rights
681-90. Quarterly 24: 237-63.

73
This content downloaded from 146.189.194.69 on Thu, 07 Jan 2016 05:38:52 UTC
All use subject to JSTOR Terms and Conditions