Weak states, human rights violations, and the outbreak of .../67531/metadc4748/m2/...Rost, Nicolas,...
Transcript of Weak states, human rights violations, and the outbreak of .../67531/metadc4748/m2/...Rost, Nicolas,...
APPROVED: Steven C. Poe, Major Professor T. David Mason, Minor Professor J. Michael Greig, Minor Professor Steven Forde, Director of Graduate Studies James Meernik, Chair of the Department of
Political Science Sandra L. Terrell, Dean of the Robert B.
Toulouse School of Graduate Studies
WEAK STATES, HUMAN RIGHTS VIOLATIONS,
AND THE OUTBREAK OF CIVIL WAR
Nicolas Rost
Thesis Prepared for the Degree of
MASTER OF SCIENCE
UNIVERSITY OF NORTH TEXAS
May 2005
Rost, Nicolas, Weak states, human rights violations, and the outbreak of civil
war. Master of Science (Political Science), May 2005, 77 pp, 4 tables, 3 illustrations,
references, 106 titles.
In recent years, explanations for the occurrence of civil war have mainly
emphasized state weakness as providing an opportunity for greed-based rebellions. Yet,
this explanation leaves many questions open, as it cannot distinguish between weak
states that do and those that do not experience civil war. In this paper, I argue that
abuses of personal integrity rights, committed or sponsored by the government, provide
this missing link. The theory is illustrated and formalized in a game-theoretic model and
then tested empirically, building on earlier work by Fearon and Laitin (2003a) and
Sambanis (2004). The results show that repression is highly significant in both statistical
and substantive terms. According to one model, the probability of civil war onset
increases by a factor of almost 16 in highly repressive countries compared to countries
with no repression. Further robustness tests across alternative civil war lists largely
confirm the importance of human rights abuses in explaining the occurrence of civil
war.
ii
ACKNOWLEDGEMENTS
I would first and foremost like to thank my committee chair Dr. Steven C. Poe
and my committee members Dr. T. David Mason and Dr. J. Michael Greig for their
mentoring with this thesis, but also with other papers during the past two years. Other
professors have offered valuable advice, especially Dr. Emily Clough. I would also like to
thank my fellow students, especially Mehmet Gurses and my office mates Chelsea
Brown and Geoff Dancy. Cece Hannah, Jerilyn Doss, and Lisa Blakeley have always
been supportive of me. I would like to thank those who have supported me financially
during the past two years: My parents, the UNT Department of Political Science, the
Fulbright Commission, International Studies Quarterly, the UNT Graduate Student
Council, and the Robert B. Toulouse School of Graduate Studies. Many thanks to my
friends here and in Germany and my family for their listening, comments, and
encouragements. Finally, my gratitude goes to Savera Kashmiri for her support
throughout the process of writing this thesis.
iii
TABLE OF CONTENTS
Page ACKNOWLEDGEMENTS...........................................................................................ii LIST OF TABLES ...................................................................................................iv LIST OF ILLUSTRATIONS ...................................................................................... v INTRODUCTION ................................................................................................... 1 LITERATURE & THEORY........................................................................................ 6
A. The Repression-Rebellion Link............................................................ 6
B. Democracy and Repression ...............................................................12
C. Weak States and Civil War Onset.......................................................17
i) Economic and Demographic Aspects of State Weakness ..17
ii) Geographic Aspects of State Weakness ..........................19
iii) Potential War Spoils......................................................19
iv) Cultural Factors ............................................................20 A GAME-THEORETICAL MODEL.............................................................................22 EMPIRICAL TESTS ...............................................................................................33
A. The Repression-Rebellion Model ........................................................33
i) The Dependent Variable................................................33
ii) Independent Variables: Repression and Democracy.........34
iii) Other Independent Variables.........................................39
B. Results............................................................................................41
C. Substantial Interpretation and Simulations .........................................46
i) The Substantial Impacts of the Independent Variables ....47
ii) Hypothetical Cases: The Four Subgames ........................50
D. Robustness Tests .............................................................................52 CONCLUSION ......................................................................................................56
APPENDIX...........................................................................................................66
REFERENCES.......................................................................................................70
iv
LIST OF TABLES
Page 1. Logit regression on the probability of civil war onset .....................................61 2. Simulated effects of statistically significant variables .....................................63 3. Logit regression on the probability of civil war onset - robustness tests ..........64 4. Logit regression on the probability of civil war onset - overall repression replaced
by political imprisonment............................................................................65
v
LIST OF ILLUSTRATIONS
Page
1. The onset of civil war in an extensive game form .........................................60 2. Civil war onset and the (lagged) level of democracy......................................62 3. Civil war onset and the (lagged and modified) PTS .......................................62
1
I N T R O D U C T I O N
Why do civil wars break out in some countries but not in others? One of the
difficulties in explaining the occurrence of civil war lies in the collective action problem –
or ‘rebel’s dilemma’ (Lichbach 1995) – that potential rebels face. In recent years, greed-
based explanations of civil war onset have received much academic attention. This class
of explanations assumes that greedy rebels act as soon as the political opportunity
structure allows it, that is, mainly in weak states. The benefits of waging a civil war
represent high selective incentives that make it an individually rational choice to initiate
an insurgency, as long as the state is so weak that it cannot easily defeat an
insurgency. Governments often use a repressive strategy in order to deter violent
rebellions from occurring, providing a motivation for the government and government-
sponsored actors to commit acts of incredible cruelty.
In this study, I argue that government-sponsored abuses of personal integrity
rights increase the probability of a civil war starting, a factor that has been overlooked
in much of the literature that focuses on political opportunity structures. Government
repression often leads to the exact opposite of what it is aimed for: It significantly
increases the probability of civil war onset rather than deterring potential rebels. While
political entrepreneurs – potential rebel elites – react to political opportunities to realize
war spoils, these selective incentives are not available to nonelites that might join the
rebels, at least not to the same extent. Especially at the beginning of a civil war, rebel
groups rarely have the means to compensate all their members. Nonelites, in contrast,
will compare the risks of participating to the costs of not participating. The government,
2
by indiscriminately abusing personal integrity rights, puts a cost on nonelites even if
they do not participate. When government repression and state weakness coexist is civil
war an especially probable outcome.
The theoretical link between state terror and the occurrence of violent rebellion
provides an important part of the explanation of when and why civil wars break out.
First, on an individual level, the repression-rebellion link contributes to an
understanding of the reasons a country’s nonelites might have to join an emerging
rebel movement. The goods a successful rebellion is expected to produce often are
largely public goods. Free-riding is then the optimal choice, as the risks of participating
are avoided, while the public good is available to everyone. Rebel movements can
promise or provide individual incentives to a selected few, but not to all its active and
passive supporters. Once the government1 starts to indiscriminately violate basic human
rights, simple nonelites – whether they oppose the government or not – are faced with
a threat, and free-riding is no longer costless. As the risk faced by simple nonelites
becomes similar to the danger they would face by joining the rebels, the individual
approaches indifference between the two choices (Mason and Krane 1989). In this
study, I conceptualize repression to be one of the most important factors in increasing
the probability of civil war onset. Yet, individuals may have other reasons to join or
support a rebel movement. These include feelings of vengeance or anger, for example,
as well as ideological commitments, social expectations, or low opportunity costs.
Furthermore, rebel organizations exercising control over an area will often try to coerce 1 Throughout this paper, I make the simplifying assumption of the existence of a government and one rebel group fighting each other. Some actual cases, however (like Somalia in the 1990s), are much more complicated.
3
nonelites into cooperation, though they can hardly implement perfect control (Gates
2002).
Second, the repression-rebellion link also provides an explanation of civil war
onset at the country level. Weak states have consistently been found to be more prone
to experience the outbreak of a civil war. Still, governments in weak states face a
strategic choice and weak states are far more common in time and space than are
violent insurgencies. This indicates that even in weak states, armed rebellions may be
avoided. Governments can opt for working towards a peaceful settlement of emerging
conflicts and try to prevent conflicts from arising outside the established political
system. This includes avoiding too much of economic inequality, institutionalizing
democracy, and respecting human rights. Though being weak, governments can try to
accommodate the opposition, and include all crucial contenders in the polity (Tilly
1978). Alternatively, they can use a strategy of repression, hoping to deter potential
rebels. This strategy, however, is dangerous and easily boomerangs, increasing the
probability of civil war instead of deterring it. The characteristics of a weak state –
economic and political discrimination, corruption, the failure of the government to
provide basic services, ‘bad governance’ – may provide both the ideological
legitimization and the practical opportunity for starting a rebellion (e.g., Collier and
Hoeffler 2001; Fearon and Laitin 2003a). But whether enough individuals join and
support the emerging insurgency depends on their alternative options: When
indiscriminate government terror imposes costs on people who lead a ‘normal’ life, their
participating becomes more likely. In other words, state weakness provides the political
4
opportunity for potential rebel leaders to realize spoils from an insurgency; government
repression affects individual preference structures in a way that makes participation a
likely choice for at least parts of society.2
If repression is such a bad choice, then why do governments so often go this
way? Civil wars are distinguished from interstate conflicts by a power asymmetry
between government and rebels (Zartman 1995). Rebel organizations therefore employ
a military strategy of hit-and-run, waging a guerilla war. By doing so, they have to rely
on popular support, and government forces then try to undercut this support by
“draining the sea” (Valentino et al. 2004; Azam and Hoeffler 2002). Where the rebels
hide in the mountains and the government chooses to use a repressive strategy,
civilians become easy and likely targets.
This explains the situation during a civil war, but why do governments repress
before a civil war starts? The more power is concentrated in a political system, the
more contending groups are left outside of the polity (Tilly 1978). In every authoritarian
regime groups are excluded from the political system and even in democracies, some
groups may be barred from power and denied representation. When governments do
not want to include contending groups and do not have other means of accommodating
them (like a high level of economic development), they will resort to repression, even at
the risk of civil war.
In the following section, I discuss the literature on civil wars, state repression,
and democracy, highlighting the theoretical linkages that have been put forward. I
2 For a discussion of the opportunity/willingness framework, see Most and Starr (1989).
5
combine arguments on greed and grievance and collective action problems into a
theoretical framework that includes state weakness and government repression to
explain potential rebel leaders’ and nonelites’ decisions to rebel. I separately discuss the
links between civil war and repression, the political system, and state weakness,
respectively. In section 2, this theory is then formulated in game-theoretic terms. Based
on the hypotheses of the theoretical framework and the comparative statics of the
game, an empirical model is constructed to test the theoretical propositions (section 3).
The results, which mostly support the theory, are interpreted, followed by a discussion
of their possible practical implications and ideas for further research.
6
L I T E R A T U R E A N D T H E O R Y
This section is composed of three sections, discussing the three aspects that
increase the probability of civil war onset: repression, characteristics of the political
system, and state weakness. I review the literature relevant to each section and
generate testable hypotheses to describe the relationship between civil war and
repression, the political system, and state weakness, respectively. Whereas state
weakness provides an opportunity for potential rebel elites to realize considerable war
spoils, nonelites are only willing to join if their costs of not joining are high, as they are
under repressive regimes. In section 2, these three parts are tied together in a game-
theoretical model, which allows distinguishing between the effects of repression in
strong and in weak states.
A. The Repression-Rebellion Link
In the past two decades, quantitative human rights research has proliferated
considerably. Typically, the vast amount of these studies has tried to explain under
what conditions human rights violations occur. Six factors have consistently been found
to impact the level of violations of personal integrity rights (Poe and Tate 1994; Poe et
al. 1999): past repression (Davenport 1995, 1996a; Richards et al. 2001), democracy
(Henderson 1991; Fein 1995; Davenport 1995, 2004; Zanger 2000; Regan and
Henderson 2002; Harff 2003; Davenport and Armstrong 2004), the level of economic
development (Mitchell and McCormick 1988; Carey 2004), population size (Henderson
1993; Carey 2004), international war, and civil war (Krain 1997; Zanger 2000; Harff
7
2003).3 The effects of all these factors, except for past repression, have been
interpreted in a theoretical framework of government strength and the perceived threat
from the political opposition (Poe 2004). In this ‘strength/threat’ framework, the
decision to violate human rights is mainly perceived as a reaction to threats posed to
the government’s legitimacy. A number of empirical studies have confirmed that
repression tends to increase when governments are faced with violent protest (Pion-
Berlin and Lopez 1991; Davenport 1995; Gartner and Regan 1996; Regan and
Henderson 2002; Carey 2004). Repression, however, often has the opposite effect of
provoking violent conflict, rather than deterring it.4
What these studies generally neglect is that protests of any form might be a
reaction to state terror. Numerous studies in the civil war literature have found that
state suppression of political rights and civil liberties, instead of deterring rebellions,
actually helps to provoke uprisings, at least at intermediate levels (Gurr 1970; Tilly
1978; Muller 1985; Muller and Seligson 1987; Muller and Weede 1990; Boswell and
Dixon 1990; Schock 1996). Studies focusing on ethnic rebellion have confirmed this
relationship (Gurr and Moore 1997). Discrimination, another form of repression, has a
comparable effect: The State Failure project found that countries with deliberate ethnic
discrimination are 13 times more likely to experience ethnic civil war as compared to
3 There has been some research on the influence of international treaties (e.g., Keith 1999; Hathaway 2002), economic aid (e.g., Regan 1995), international trade and FDI (e.g., Apodaca 2001), constitutional provisions for human rights (e.g., Davenport 1996b; Keith 2002; Keith and Poe 2004), and ethnic diversity (e.g., Walker and Poe 2002), but findings are rather mixed. 4 Most of these findings, as well as this study, are based on a subset of human rights, the right to personal integrity, which prohibits political imprisonment, torture, ‘disappearances,’ and killings. Some quantitative studies on other subsets of international human rights have been conducted (e.g., Poe et al. 1997; Apodaca 1998), but mostly the discourse has been theoretical or based on case studies.
8
countries with no or little discrimination (Goldstone et al. 2000, 35).5 Some of these
studies have found that ‘semirepressive’ regimes are most civil war prone. The question
is whether government repression provokes or deters violent rebellion. Lichbach (1987)
argues that opposition movements respond to government repression by switching to
alternative strategies. Gupta et al. (1993) find that the effects of government sanctions
vary with the regime type of the country (see also Moore 1998).
Repression, in most of these studies, is either operationalized by using Gastil’s
Freedom House indices of political rights or civil liberties, or by measuring government
sanctions. In this study, I define repression very differently, as human rights abuses
committed or sponsored by the government.6 As I explain below, this latter definition of
government repression has a different theoretical impact on civil war onset, operating
through different causal mechanisms. It further allows a more clear-cut distinction
between democracy and repression (note that the Freedom House indices have
oftentimes been used as indicators of democracy as well). Whereas the citizens in an
authoritarian country may collectively prefer a more open and participatory political
system, it becomes individually rational for nonelites suffering from a repressive regime
to engage in violent rebellion.
The literature on inequality and relative deprivation (e.g., Davies 1962; Gurr
1968; Muller and Seligson 1987; Midlarsky 1988) assumes that economic grievances
lead to violent domestic conflict. The problem with this argument, as pointed out by
5 Fearon and Laitin (2003, 85), on the other hand, find no such effect. 6 My measure of human rights abuses, the modified Political Terror Scale, and Freedom House’s measures of political rights and civil liberties are correlated with a factor of .50 and .56, respectively. From here on, ‘repression’ refers to government-sponsored abuses of personal integrity rights.
9
many scholars, is to explain how the collective action problem is overcome. A higher
average income might be desirable to many, but most prefer to free-ride instead of
shouldering the costs of achieving such a change. While low income reduces the
opportunity costs of rebelling, indiscriminate government repression places costs on
nonelites, even if they do not participate. As a consequence, repression makes it
individually and not only collectively more desirable to join or support the rebels.
On the individual level, Mason and Krane (1989) develop a rational choice model
that lays out how repression creates a conflict spiral and drives civilians into joining the
rebels. They thereby provide an explanation of one important means of overcoming the
collective action problem political movements are faced with (Olson 1965). Other
solutions offered in the literature include selective incentives (Olson 1965; Oliver 1980;
Moore 1995) market, community, contract, and hierarchy (Lichbach 1995, for
community see also Taylor 1988), psychological and reputational incentives (Chong
1991), leadership (Van Belle 1996), and framing processes (Berejikian 1992). Mason
and Krane’s (1989) explanation differs from the ones above in that the incentive to join
the insurgents is not created within the group but externally by the government.
Here, it becomes important to distinguish between potential rebel elites who
initiate an insurgency and nonelites who are willing to join the guerrilla or not (Mason
and Krane 1989, 176). The elites can expect to gain considerable benefits from a
successful rebellion, political or economic. These selective war spoils, however, can
rarely be extended to the majority of participants so that nonelites7 can only realize
7 The rebel elites are nonelites too, of course. Here, I use the word ‘nonelites’ to refer to all habitants of a country except government agents and rebel elites.
10
small material benefits and possibly psychological or reputational ones. Henceforth,
nonelites are willing to join or support the rebels if their costs of not joining are high,
i.e. if the government indiscriminately commits human rights violations. Indiscriminately
does not necessarily mean that the whole population is affected, but merely that the
government does not distinguish between actual rebels and nonelites from, for
example, the same region or the same ethnic, religious, socio-economic, or political
group. Nonelites would prefer to stay out of the conflict (Mason and Krane 1989, 176),
but government repression drives them into rebellion.8
Hypothesis 1: The more repression a government exerts, the higher the probability of a civil war to break out.
Some might question whether repression increases the probability of all kinds of
civil wars. Collier and Hoeffler (2000) distinguish between greed- and grievance-based
civil wars which raises the concern of whether the probability of greed-based civil wars
increases to the same extent as the probability of grievance-based civil wars in weak
states. Empirically, though, rebel movements often shift from one type to the other or
combine aspects of both. The rebel leadership might be driven by greed, whereas the
rank and file members are motivated by grievances. In greed-based civil wars,
insurgents seek to profit from the fighting while it is still ongoing, for example by
trafficking in contraband (Fearon 2004) or looting. This type of insurgency, if victorious,
generates additional amounts of selective incentives, at least for the elites, making it
easier to overcome the rebel’s dilemma. Greed-based rebellions might be seen as
business or investment (Collier et al. 2004). 8 This argument should not be interpreted as a normative excuse or legitimization for violent behavior on an individual level.
11
In grievance-based insurgencies, on the other hand, the civil war is only
interpreted as an investment (Collier et al. 2004). Although the rebels may have to rely
on creating selective incentive structures to attract members, the emphasis is put on
producing a collective good (e.g., an independent country, a new government). Many
rebellions start out as grievance-based insurgencies, but then turn into greed-based
ones. In Peru, for instance, the Sendero Luminoso was an ideological movement in the
beginning, until the rebels discovered the potential profits from growing and smuggling
drugs. Fearon (2004) finds that if the rebels use contraband as a financial source, the
duration of civil wars increases considerably. As a result, the distinction between greed-
and grievance-based civil wars should be more important for studies on war duration
than studies on onset.
Further, I argue that greed should play a role for potential rebel elites, more
than for the nonelites that join them. Elites can usually realize substantial spoils in the
war, and may react to openings in the political opportunity structure by initiating
insurgencies. Nonelites, on the other hand, are willing to join only if the danger of
participating is lower than the costs of not participating. Participants in a rebellion face
the dangers of death or lifelong imprisonment, but under repressive regimes, non-
participants as well face similar risks of being arbitrarily imprisoned, tortured,
‘disappeared,’ or killed. Even if an insurgency is purely greed-based from the beginning,
nonelites will only join if alternative payoffs from not joining are low, as the rebels
cannot provide sizable incentives for too many members. While few civil wars start off
12
as being greed-based, I expect repression to play an important role even in these rare
cases.
A potential rebel elite is of course aware of this phenomenon and will consider
two aspects before making the decision to start a civil war. First, they will evaluate the
probability of success in their struggle, or at least, the probability of surviving long
enough to realize some spoils from the insurgency. Zartman (1995, 9) quotes Kissinger
stating that the “guerilla wins if he does not lose; the conventional army loses if it does
not win.” According to this line of thought, potential rebel leaders do not necessarily
have to expect to win outright before mounting an insurgency, but they do have to
believe that they will not lose in order to realize some spoils. For this calculation, mainly
factors of state strength play a role, as in Fearon and Laitin’s (2003a) argument. But
there is a second, if related, factor: A potential rebel elite will have to reflect on
whether they can gain enough support from the population, both by attracting new
members and receiving enough active and passive support from the population. One
main reason for whether the rebels will succeed is the level of government-sanctioned
terror that nonelites face, i.e. the costs of not joining or supporting the rebels.
B. Democracy and Repression
In this section, I develop theoretical expectations on the effects of democracy on
civil war onset, once repression is controlled. While repression is usually reduced as the
level of democracy increases (Henderson 1993; Poe and Tate 1994; Fein 1995;
Davenport 1995, 2004; Poe et al. 1999; Zanger 2000; Regan and Henderson 2002;
13
Harff 2003; Davenport and Armstrong 2004), democracy and non-repressiveness are
conceptually and empirically distinct phenomena.
Many of the studies that explore whether repression increases the probability of
rebellion use the Freedom House 7-point measure of civil liberties to model repression
(e.g., Muller 1985; Muller and Seligson 1987; Muller and Weede 1990; Boswell and
Dixon 1990; Schock 1996), where ‘1’ represents general respect for civil liberties, and
‘7’ wide-spread suppression of these liberties. Political rights refer to the electoral
process, political pluralism and participation, and the functioning of the government.
Civil liberties refer to the freedom of expression and belief, associational and
organizational rights, rule of law, personal autonomy and individual rights.9 The
problems with this measure are obvious: Civil liberties may be suppressed without
widespread violations of the most basic human rights. Singapore in 1985 and 1986, for
instance, was assigned a ‘5’ on the civil liberties index, but almost no violations of
personal integrity rights occurred.
The relationship between repression and democracy has received much scholarly
attention. While Poe and Tate (1994) found a linear influence (the more democratic a
country is the fewer human rights abuses), Fein’s (1995) ‘More Murder in the Middle’
hypothesis posits that more violations should occur in so-called anocracies. These
countries that are neither fully consolidated democracies nor fully established
autocracies, are expected to experience more repression because governments are too
9 http://www.freedomhouse.org/research/freeworld/2003/methodology.htm, accessed March 2005.
14
weak to effectively deter the opposition and, at the same time, they do not have the
means to politically or economically accommodate opposition demands.
Davenport and Armstrong (2004) explain that the relationship between
democracy and repression is even more complex than Fein’s inverted u-curve. They
develop a threshold-model and find that the level of democracy has no impact on the
level of repression up to a relatively high threshold. Only beyond that level has
democracy an alleviating impact on human rights violations. In other words, there are
about as many authoritarian states and anocracies that repress human rights as there
are that do not. Moreover, fully developed democracy seems to be a sufficient condition
for avoiding very high levels of repression. In their sample, not a single fully developed
democracy experienced the highest levels of repression (there has been one case since,
Israel in 2002 and 2003), and only a couple show scores of intermediate to high levels
of human rights abuses.
The implications of this complex relationship between democracy and repression
for the study of civil war onset are potentially important. Fearon and Laitin (2003a)
have not found democracy to have a significant impact on the likelihood of a civil war to
break out, though a dichotomous measure of anocracy was significant, as well as a
measure of political instability. Sambanis (2004) does not include a democracy
measure, but only the anocracy variable in his models, and finds it of only moderate
influence.
15
Since authoritarian regimes can choose whether or not to repress, it is important
to control for both repression and democracy in a multivariate model.10 While
repression is expected to impact the probability of civil war onset independently of the
level of democracy for the reasons laid out above – the repression-rebellion link – I also
expect democracy to significantly decrease this probability. Democracies do not oppress
political conflicts but provide rules to peacefully solve these conflicts within the political
system. This makes it easier to find compromise solutions and to share political power
across different ethnic or ideological groups in a country. Thereby, fewer groups are left
outside the polity and have an incentive to violently claim inclusion. A higher political
adaptability of democracies may also make it easier to solve conflicts nonviolently
(Lacina 2005). In addition to democratic institutions, democratic norms exclude violent
rebellion as a legitimate option.11
Hypothesis 2a: Once repression is controlled, democracy will exert a significant impact on the probability of a civil war to break out, decreasing this probability.
Hegre et al. (2001) find support for a domestic democratic peace, but they find
that both anocracies and politically unstable countries are more conflict prone. Similarly,
Fearon and Laitin (2003a) and Sambanis (2004) find significant effects for anocracy and
political instability in some of their models. Anocracies are more civil war prone because
neither are democratic institutions fully established, nor has an authoritarian regime
consolidated its power. Anocracies have oftentimes unstable, unconsolidated political
systems. Several groups may compete for control of the government while the rules of
10 Or, in Muller and Seligson’s (1987, 432) and Muller and Weede’s (1990, 647) words, for both structural and behavioral repression. 11 However, I cannot empirically discriminate between these alternative explanations.
16
the game have not been institutionalized. Similarly, political instability delegitimizes the
ones in power and shows that they may be vulnerable.
The same should be especially true in newly independent states. Sambanis
(2004, 837), though, criticizes the use of the ‘new states’ variable in Fearon and Laitin’s
(2003a) study, since it might be correlated with some other variables, especially
instability, anocracy, income, and ethnic fractionalization. The state weakness
argument, however, leads to the expectation that new states are more war prone. In
Fearon and Laitin’s (2003a, 84) Model 1, ‘new state’ was the substantially and
statistically most significant dichotomous variable. Further, new states have been found
to be more likely to experience other forms of political instability, including interstate
disputes (see Brecher et al. 2000; Carment 1993) and genocides and politicides (Krain
1997, 346; Harff 2003, 62).
Finally, military regimes are expected to increase the probability of civil war
onset. Military regimes are defined as any regime with a military person as the chief
executive that comes into power following a successful military coup (Madani 1992).
Military regimes can be expected to represent politically weak governments. They have
installed themselves by staging a coup and rarely enjoy widespread popular support.
Further, the very fact that a coup was staged indicates that government structures
were weak in the first place.
Hypothesis 2b: Civil wars are more likely to break out in anocracies.
Hypothesis 2c: Countries with a recent history of political instability show a higher probability of experiencing a civil war.
Hypothesis 2d: Civil wars are more likely to break out in countries governed by military regimes.
17
Hypothesis 2e: Civil wars are more likely to break out in newly independent countries.
These four hypotheses lead to the next section which discusses state weakness:
All four factors are signs of political state weakness, while in the next section I discuss
economic, demographic, and geographic aspects of state weakness.
C. Weak States And Civil War Onset
State weakness has long been associated with the occurrence of civil war (e.g.,
Skocpol 1979). State weakness produces the openings in the political opportunity
structure that can be seized upon by initiating an insurgency. State strength, in this
theoretical concept, mainly effects the calculations of potential rebel leaders. They see a
greater chance for realizing benefits from an insurgency the weaker the state is,
regardless of whether they perceive rebellion as investment or business. If the state is
strong enough, an insurgency will be crushed soon, leaving no room for any benefits to
materialize. State weakness thus creates an opportunity for rebellion (Collier and
Hoeffler 2001), but it does not necessary create the willingness of nonelites to join or
support the movement.
i) Economic and Demographic Aspects of State Weakness
In Sambanis’ (2004, Table 4) study, the only two variables that show a
statistically significant impact across 12 different lists of civil wars, 1960-1993, are GDP
18
per capita and the log of population size (see also Elbadawi and Sambanis 2002).12 Both
a very low level of economic development and a large population produce relatively
weak states and make it easier for the political opposition to organize and start in a
violent rebellion.13 A low economic standing makes it more complicated for a
government that is not willing to share power to accommodate contending groups.14 In
addition, Mason and Krane (1989) argue that governments that are neither willing to
launch redistributive programs to accommodate opposition demands, nor have the
economic resources to do so, are likely to switch to a repressive strategy. A large
population size, on the other hands, makes it harder for governments to control the
whole country. This also increases the probability of existing contending groups that
feel politically or economically underrepresented. In addition, both these conditions
have consistently been found to be associated with relatively high levels of repression,
in turn increasing the probability of a civil war (Mitchell and McCormick 1988;
Henderson 1993; Poe and Tate 1994; Poe et al. 1999; Carey 2004).
Hypothesis 3a: The level of economic development is negatively correlated with the probability of a civil war breaking out.
Hypothesis 3b: A country’s population size is positively correlated with the probability of a civil war breaking out.
12 With one exception: population size was not significant with Regan’s list of civil conflicts. Both per capita income and population size, as well as all other independent variables in Sambanis’ models, are lagged by one year. Both are also significant over three different model specifications, four different civil war lists, and the 1945-1999 period. 13 Sambanis (2004, 844) argues that the results for population size might be an artifact of the high absolute threshold of deaths used to code civil wars. Indeed, with Regan’s civil conflict data, where a lower threshold of 200 deaths is used, the coefficient for population size is not significant. However, the state strength argument provides some reasons to theoretically expect more civil wars in larger countries, as discussed below. 14 Fearon and Laitin (2003, 76) use the level of economic development also as a proxy for the quality of counterinsurgency measures. They contend that “weak local policing or inept and corrupt counterinsurgency practices […] often include a propensity for brutal and indiscriminate retaliation that helps drive noncombatant locals into rebel forces.”
19
The level of economic development or the average income also impacts the
opportunity costs of participating in an upheaval. The higher the opportunity costs, the
less likely nonelites will be to join or support the insurgents. It becomes thereby harder
for rebel elites to act as political entrepreneurs and to recruit followers and attract
supporters. As potential rebel elites are aware of this difficulty, they are less likely to
initiate a rebellion.
ii) Geographic Aspects of State Weakness
In addition to economic and demographic aspects, geographic aspects of a
country may contribute to state weakness. Specifically, rough terrain makes it easier for
rebels to hide from government troops, to wage a guerilla style war (Collier and
Hoeffler 2001; Fearon and Laitin 2003a), and to traffic contraband.
Hypothesis 3c: Countries with rough terrain will show a higher probability of experiencing a civil war.
iii) Potential War Spoils
Fearon and Laitin (2003a) find that civil wars are more likely to start in states
with a high amount of oil exports. Oil exports are not as much a sign of state weakness
as they represent potential spoils for the rebels should they succeed in overthrowing
the government or creating their own state in an oil rich area. Oil fields can usually not
be exploited by the insurgents during a civil war. Exploitation requires exalted
investment and foreign companies will be turned away by the political and economic
instability of that country. The government that controls oil fields, on the other hand, is
20
in a privileged situation as it can easily exclude large parts of the population from the
economic gains. While oil revenues may encourage rent-seeking and corruption among
government elites, these revenues can also be used to battle the insurgents. If the
rebels succeed, however, large benefits from the oil production materialize. This
increases the incentives for potential rebel leaders to stage a rebellion, but not
necessarily for nonelites, as they cannot expect to receive much of these benefits.
Accordingly, I expect the following hypothesis to hold true.
Hypothesis 3d: Oil exporting states will show a higher probability of experiencing a civil war.
iv) Cultural Factors
While neither Fearon and Laitin (2003a) nor Sambanis (2004) find much
evidence for an increasing civil war probability with rising levels of ethnic and religious
fractionalization, it is important to control for these cultural factors, as many civil wars
are fought over ethnic or religious issues (Kaufmann 1996). Even if ethnicity or religion
is not the primary cause, both government and opposition elites oftentimes
instrumentalize and exploit, or even create, such cleavages (Horowitz 1985; Kaufman
1996). Ethnic and religious cleavages also make it easier for rebel leaders to recruit new
members and retain them as they can concentrate on specific groups in society (Gates
2001). Lake and Rothchild (1996) suggest that ethnic wars arise from information
failures and problems of credible commitment, which should be more likely to occur in
weak states. Fearon and Laitin (1996), however, note that ethnic cooperation occurs far
more often than ethnic conflict; Mueller (2000) even questions the viability of ethnic
21
explanations for civil war. Ethnic and religious fractionalization are included in the
empirical model to test the following two hypotheses:
Hypothesis 3e: Countries with a high level of ethnic fractionalization will show a higher probability of experiencing a civil war.
Hypothesis 3f: Countries with a high level of religious fractionalization will show a higher probability of experiencing a civil war.
To summarize, this theory distinguishes between the opportunity and willingness
to start a civil war. Willingness does not only include the opportunity costs of rebelling,
but also the costs of not joining or supporting the rebels. The theory treats potential
rebel leaders and nonelites separately. Elites perceive state weakness as an opportunity
to start an insurgency, but they depend on a substantial number of nonelites to join or
support the insurgency group. Nonelites are willing to join if the costs of not joining are
high, i.e. if the government indiscriminately commits human rights abuses. In the
following section, a game-theoretic model is constructed to illustrate and formalize the
theory, followed by an empirical test.
22
A G A M E – T H E O R E T I C A L M O D E L
The game described in this section illustrates the theoretical framework outlined
above. From the expected utility functions of potential rebel elites and nonelites,
comparative statics are calculated, which correspond to the hypotheses derived from
the theory. Formalizing a theoretical argument helps to ensure internal logical
consistency. In this case, the game-theoretical model also allows examining different
combinations of the factors increasing the probability of civil war onset. For example,
what impact does repression have, contingent on the level of state weakness?
Comparative statics allow specifying the theoretically expected impact of each factor,
holding other influences constant. These effects can then be tested empirically. From
the formal model below, some additional theoretical expectations are generated. To
give a short summary, civil war is only found to be the equilibrium outcome when state
weakness and state terror coexist. In addition, the size of potential war spoils, and the
opportunity costs of rebelling should also have an effect on civil war onset. Finally, rebel
elites should be more sensitive to changes in any of these aspects compared to
nonelites.
The game tree is depicted in Figure 1. Nature moves first, choosing between a
strong and a weak state. This move is common knowledge to all other players, as the
whole play is one of perfect information. The government moves next, deciding
between a repressive and a non-repressive strategy. Third, the opposition elites (or
extremists) decide whether or not to mount a rebellion and fourth, the nonelites (or
23
moderates) decide whether or not to join the movement. 15 Although in the sequence of
this model, potential leaders first decide whether to mount a rebellion and nonelites
later decide whether to join the movement, reality is much more complex. Opposition
elites, the population at large, and the government interact over an extended period of
time before it comes to a civil war.
In the game tree, state weakness and repression are modeled as dichotomous
choices. In the comparative statics section below, this unrealistic assumption is relaxed
to produce testable hypotheses. Government, in this model, does not necessarily refer
only to the party in power. In a democratic system, even in a weak state, this might
include political opposition parties that are sufficiently tied to the polity and can either
expect to win elections at some time in the future, be part of a government coalition, or
exert political influence regardless of their opposition status. Even in democracies,
however, groups are sometimes excluded from the polity.
For each combination of strong/weak states and repression/no repression, there
are three possible outcomes. These four combinations denote the four subgames of the
game tree. If opposition elites do not start a rebellion, nothing happens. If they do
rebel, on the other hand, the nonelites decide to join the movement or not. If they do
join, a civil war breaks out, as the rebel movement becomes large enough. If they do
15 The model applies equally well if the word ‘elite’ is replaced by the word ‘extremists,’ and ‘population’ or ‘nonelites’ by ‘moderates.’ Elites are similar to extremists in that they are the first to decide whether to mount a rebellion; moderates are similar to nonelites in that they later decide whether to join or not. Further, elites and extremists can expect higher potential benefits from the insurgency, albeit for different reasons. While political-economic payoffs might be more salient to political entrepreneurs, extremists profit from political-ideological benefits.
24
not join, the movement will either break down, or end up as a terrorist group or an
assemblage of criminal gangs.
In order to calculate potential elites’ and nonelites’ payoffs, I construct expected
utility functions for both players. State weakness and repression have an impact on
both potential rebel elites’ and nonelites’ calculations. Potential elites react to
opportunities posed by the weakness of the state. They can benefit from criminal or
terrorist activities, as well as from the spoils of a civil war if the state is sufficiently weak
so that neither is crushed immediately. Potential rebel elites prefer, however, civil war
as they can then share the risk of violently opposing the government with the
participating nonelites. There are many beneficial outcomes for potential rebel leaders:
They may become the new political leaders in a newly formed state after secession or if
the central government is thrown over (insurgency as investment); they may make
profits during the conflict, e.g., from looting, drug trafficking or blood diamonds
(insurgency as business, Collier et al. 2004); or they may realize economic or political
profits from criminal or terrorist activities. These spoils are assumed to exceed the
benefits of leading a ‘normal’ life. For potential rebel leaders, therefore, opportunity is
more important than willingness, as they are always willing to initiate an insurgency.
For nonelites, on the other hand, joining a rebel movement is always risky, as it
involves the danger of capture or death and, in most cases, the abandoning of their
families and jobs. Even a lower level of support – like hiding rebels or providing them
with food or merely not denouncing them – involves risks. While nonelites might turn to
the rebels hoping that they will shelter them from indiscriminate government repression
25
(Mason and Krane 1989), supporters of the rebels will always face the potential danger
of getting caught or killed by government agents, at least as long as the state is not
perfectly weak. They do receive some selective benefits from joining the rebels, but
these will be very low in material terms and not high enough in social (reputational) and
psychological terms to rationally justify participation. Since benefits are low, nonelites
are motivated by willingness more than by opportunity. Specifically, they will compare
the costs of joining to the costs of not joining. These costs mainly stem from
indiscriminate human rights abuses committed by the government against the
population. Either way, if the costs of not joining are sufficiently high to outweigh the
danger of joining, it becomes rational for the individual to join.
The expected utility function for nonelites (C) is as follows:
E(U)C = j [wm – (1 – w) d] + (1 – j) [n – r], (1)
where j is the decision to join a rebel movement and 1 – j is the decision to abstain (j = {0, 1});
w is state weakness (0 ≤ w ≤ 1), 1 – w, therefore, is state strength;
m is the motivation to join the rebels (selective material or reputational benefits, but also irrational feelings of vengeance, ideological reasons, etc.; 0 ≤ m ≤ 1);
d is the danger associated with joining the rebels (d = 1, since it might involve death or lifelong imprisonment);
n are the economic, social, and psychological benefits of leading a ‘normal’ life, i.e. of not joining the rebels (0 ≤ m < n ≤ 1);16 and
r is government repression (0 ≤ r ≤ 1). 17
16 All the variables in this model, but especially m and n should be interpreted as individually perceived payoffs that vary across different people. 17 I decided not to discount repression by state weakness as especially politically and economically weak states seem to experience relatively high levels of government-sponsored human rights abuses.
26
If the individual joins (j = 1), she will realize her material and reputational
benefits m to an extent dependent to the weakness of the state w.18 However, these
incentives are assumed to be small, so that the danger of participating (dependent on
state strength, 1 – w)19 will make the total payoff negative in most cases.
Yet, if she does not join (j = 0), she will receive the benefits of leading a ‘normal’
life (n) but will have to bear the costs of repression (-r). If these are higher than wm –
(1 – w) d, a higher – if still negative – utility can be realized from participating. The
benefits of leading a ‘normal’ life (i.e. the opportunity costs of rebelling) are assumed to
be higher than the benefits from rebelling as a consequence of the collective action
problem or rebel’s dilemma. m refers to selective incentives only, benefits from possible
collective goods are equal for participants and non-participants and are therefore not
included in the game. The decision to join the rebels could also be modeled as a
decision of supporting the rebels to varying degrees, from passive sympathizing to
actual participation in the fighting. j would then be a continuous, rather than a
dichotomous variable.
The expected utility for potential rebel elites (E) is as follows:
E(U)E = i [ws – ((1-w) d - p)] + (1 – i) [n – r], (2)
where i is the decision to initiate an insurgency and 1 – i is the decision not to (i = {0, 1});
w, again, is state weakness (0 ≤ w ≤ 1), 1 – w is state strength;
s are the spoils that could be realized from mounting an insurgency (0 ≤ m < n < s < 1);
18 In other words, state weakness indicates the probability with which these benefits will be realized. 19 Similarly, state strength indicates the probability with which the danger materializes.
27
d is the danger associated with initiating an insurgency (d = 1), as above;
p is the rate of nonelites’ participation in the insurgency (0 ≤ p ≤ 1);
n, as before, refers to the economic, social, and psychological benefits of leading a ‘normal’ life, i.e. of not initiating an insurgency (0 < n ≤ 1); and
r is government repression (0 ≤ r ≤ 1).
If the potential leaders initiate an insurgency (i = 1), spoils s will materialize to
the extent that the state is weak.20 They will have to bear the danger d (as a function
of state strength, 1 – w), but for them, p, the rate of participation, is subtracted from
the danger, because they can share the risk if others join. On the other hand, if they do
not initiate an insurgency (i = 0), elites will have to face the same level of repression r
as the nonelites, and receive similar benefits n.
Participation, in turn, depends on how many nonelites choose to join (j = 1).
E(U)C depicts the expected utility calculation of one individual only, whereas p
expresses the rate of participation, as a result of the aggregation of many individual
choices. Participation, therefore, may be calculated from equation (1) as:
p = wm – (1-w) d + r – n; (3)
where m is now the average perceived motivation to join, over all nonelites, and n
represents the average perceived benefits of leading a ‘normal’ life, over all nonelites.
Thus, p is positive if wm + r > (1-w) d + n. Practically, p cannot be smaller than
0, although it can be mathematically. In other words, nonelites will join a rebel
movement to the extent that costs from repression (plus small selective benefits m)
exceed the danger of fighting (discounted by state strength, plus the benefits of a
20 Once more, state weakness indicates the probability with which spoils are realized.
28
‘normal’ life n). By replacing p, the complete equation for potential elites’ expected
utility then becomes:
E(U)E = i [ws – ((1-w) d – (wm – (1-w) d + r – n))] + (1 – i) [n – r]. (4)
They will then initiate a rebellion, iff:
ws – ((1-w) d – (wm – (1-w) d + r – n)) > n – r;
ws – n > (1 – w) d – p – r; or:
ws + p + r > (1 – w) d + n. (5)
Potential elites will therefore initiate an insurgency to the extent that the state is
weak, spoils are high, participation is expected to be high (as a result of a weak and
repressive state), and repression is high. Even if participation is relatively low (because
repression is low) but the state is sufficiently weak, the expected spoils may outweigh
the danger, and an insurgency will be initiated. In this case, however, few people join
and the insurgency will turn into a terrorist or criminal movement. Since the whole
game is modeled as one of perfect information, potential leaders know about how many
people will join. Still, they might start an insurgency, if they can make profit from doing
so, even if they have to bear all the danger.
If the individual choices to initiate a rebellion by potential leaders are aggregated
from equation (4) to a measure q (q like quarrel), in analogy to the participation rate p,
q = ws – {(1 – w)d – [wm – (1 – w)d + r – n]} + r – n, or:
q = ws – 2d + 2wd + wm + 2r – 2n, (6)
29
where, as before, m is now the average perceived motivation to join, over all nonelites
and potential rebel leaders, and n represents the average perceived benefits of leading
a ‘normal’ life, over all nonelites and potential leaders.
The distribution of payoffs, resulting from the expected utility equations, is
relatively simple. I have left out the payoffs for the government, since I do not
empirically test for the reasons of why human rights abuses are committed by the state
in the first place; the inclusion of government payoffs would also make the game much
more complicated. Basically, however, the government will sometimes choose to pursue
a repressive strategy and sometimes not, both in a strong and weak state. From the
point of view of potential rebel elites and nonelites, then, nature’s and government’s
move can be regarded as external, so that I can concentrate on their payoff structures.
The game tree in Figure 1 depicts the moves by nature and government, elites and
nonelites as dichotomous choices. Nature chooses state weakness w to be either 0 or 1;
similarly the government chooses between no or maximum repression so that r = {0;
1}. The payoffs at the final nodes, henceforth, are calculated by setting w, r, i, and j to
either 0 or 1. The first payoff always denotes the expected utility of the potential rebel
leaders and the second that of the nonelites.
Calculating the subgame equilibria for each of the four subgames gives the
following equilibria: (1) If the state is strong and the government does not repress (the
lower left corner in Figure 1), nonelites clearly do not join, since n > -d. Similarly,
potential elites prefer a ‘normal’ life to both civil war and terrorist/criminal activity,
because n > 1 – d > -d (d = 1). Further, they know that they would not be joined by
30
nonelites so they simply compare n to –d and choose not to initiate an insurgency. (2)
In a strong and repressive state (upper left corner), potential elites again prefer doing
nothing to mounting a rebellion that nobody joins in, since n – r > -d (r = d = 1, n >
0). While they would prefer a civil war to doing nothing (as long as n < 1), they know
that the nonelites would not join, as, again, n – r > -d. The equilibrium, therefore, is
that no movement will be initiated. (3) In a weak, but not repressive state (lower right),
nonelites still prefer not to join, as by assumption n > m. Here, however, the elites
engage in violent activity, even if they are not joined, since the spoils are larger than
the payoffs from doing nothing (s > n, by assumption). They would still prefer a civil
war, since s + p > s, but they do not depend on participation in a perfectly weak state.
Consequently, weak states should see increased criminal and/or terrorist activities
(though I do not test this hypothesis empirically). (4) In a weak and repressive state
(upper right), finally, nonelites will prefer to join an emerging insurgency, as m > n – r.
While potential elites would rebel either way (s > n – r), it is only in this scenario that
they can start a civil war, which they prefer (s + p > s). Only in a weak and repressive
state is the equilibrium a civil war.
On the aggregate level, comparative statics denote the effect of all factors on the
participation rate p and the rate q at which potential rebel elites initiate insurgencies.
The other variables (w, m, d, r, n, s) may assume any value between 0 and 1, subject
to the constraint that s > n > m. Using comparative statics, the following partial
derivatives can be derived from equation (3):
31
δp / δw = m + d > 0
δp / δm = w > 0
δp / δd = w – 1 < 0 if w < 1 and = 0 if w = 1
δp / δr = 1 > 0
δp / δn = -1 < 0
These comparative statics denote the direction of the effect of the variables in
the model on the participation rate p of the nonelites. Similarly, the effects on q, the
rate at which potential rebel leaders try to initiate rebel movements – or criminal or
terrorist groups at least – can be calculated from equation (6):
δq / δw = s + 2d + m > 0
δq / δs = w > 0
δq / δd = 2w – 2 = w – 1 < 0 if w < 1 and = 0 if w = 1
δq / δm = w > 0
δq / δr = 2 > 0
δq / δn = -2 < 0
Interestingly, but consistent with the game-theoretic model, potential leaders
react more sensitively to changes in state strength, repression, and the benefits of
leading a ‘normal’ life. This results from the fact that these factors influence q both
directly and indirectly via p. They are more likely to rebel, sometimes forming criminal
or terrorist groups, even if not joined by the nonelites. More important, the effects of all
factors are in the same direction for elites and nonelites, allowing a test of both aspects
in a single empirical model. State weakness, potential war spoils, and repression
32
increase the likelihood of civil war; state strength and the opportunity costs of rebelling
reduce the likelihood.
These comparative statics easily translate into the testable hypotheses outlined
above. m and d are assumed to be constant. State weakness is operationalized by the
level of economic development, anocracies, political instability, military regimes, new
states, population size, and the percentage of mountainous terrain. Economic
development, on the other hand, also affects n, the opportunity costs of joining the
rebels. The potential spoils s of a successful uprising are (imperfectly) operationalized
by oil exports, which provide successful guerrillas with a stable source of income. When
the insurgents succeed in overthrowing the government, they can realize the rents
associated with oil production. Finally, repression is operationalized as state-sponsored
human rights abuses, including political imprisonment, torture, ‘disappearances,’ and
killings. The operationalization of the variables is discussed in greater detail below. In
the following section, I construct an empirical model to test the theoretically derived
expectations and describe the operationalization of the variables.
33
E M P I R I C A L T E S T S
A. The Repression-Rebellion Model
The dependent variable in this research design is a dichotomous variable
denoting the beginning of a new civil war: civil war onset. I therefore use Logit
maximum likelihood estimation to estimate a multivariate model. All countries for which
data are available are included. Data on abuses of personal integrity rights are available
from 1976 on. Due to the one-year lag of this variable, the time frame of this study is
from 1977 to 1999, the last year which many civil war lists report. Countries with a
population of less than half a million people in 1990 are excluded, as they are not
included in Fearon and Laitin’s (2003) dataset, which serves as a source for many
variables.
i) The Dependent Variable
There are many different lists of civil wars, based on different operational
definitions. Even with the same definition, scholars often do not agree on whether a
specific event represents a civil war or not, since data on battle deaths, ceasefires,
military movements, etc. are often very imprecise. And if they do agree, it is even
harder to concur on the exact start and end dates. Sambanis (2004) describes these
problems in great detail and argues that models of civil war onset, prevalence, and
duration should be tested for robustness across different civil war lists. He claims that
this is especially important with variables like GDP growth, which is itself (negatively)
influenced by an erupting civil war. The same is true for human rights violations, which
34
have been found to increase during civil wars (Poe and Tate 1994; Krain 1997; Poe et
al. 1999; Zanger 2000; Harff 2003). In other words, if a list codes the start of a civil
war ‘too late,’ then the observed human rights abuses in the preceding year might be a
consequence of the war, and not one of its causes. Therefore, I test my model across
six of the twelve civil war lists in Sambanis’ (2004) dataset (see Table A1 in the
appendix). I use his ‘a-versions,’ excluding country-years during an ongoing civil war.
First, I test my model with Sambanis’ (2004) civil war codings and describe the results
in detail. I chose Sambanis’ civil war list as it is one of the most recent and well-
documented efforts to compile data on civil wars in the post-1945 era.21 Then, to test
the robustness of the results, I replace his codings with codings from the most widely
cited studies and datasets on civil war: the COW (Sarkees and Singer 2001) and
Uppsala (Gleditsch et al. 2001) datasets, Fearon and Laitin (2003a), Regan (1996), and
an extended version of Doyle and Sambanis (2000). Regan, in contrast to other studies,
uses a relatively low threshold of 200 battle deaths which allows for further robustness
checks. Similar to the correlation coefficients reported in Sambanis (2004, 834), the
different civil war lists are correlated with factors ranging from .54 to .90, for the years
1977 to 1999.
ii) Independent Variables: Repression and Democracy
In a first set of models, human rights violations are measured on the 5-points
Political Terror Scale (PTS, originally developed by Michael Stohl, see Poe and Tate
21 See the detailed case descriptions available at http://www.yale.edu/unsy/jcr/jcrdatadec04.htm, accessed March 2005.
35
1994; Poe et al. 1999). The updated version of the Political Terror Scale is available
from Gibney (2004). The scale ranges from ‘1,’ where repression is almost but not
necessarily entirely absent, to ‘5’ where state terror is inflicted indiscriminately on the
whole population. Examples of countries coded ‘1’ in 2003 include Burkina Faso,
Canada, and Kazakhstan. Examples of countries coded ‘5’ include Algeria, Colombia,
and Indonesia. Countries are coded according to the following coding scheme:
1 - Countries...under a secure rule of law, people are not imprisoned for their views, and torture is rare or exceptional...political murders are extremely rare.
2 - There is a limited amount of imprisonment for nonviolent political activity. However, few persons are affected, torture and beating are exceptional... political murder is rare.
3 - There is extensive political imprisonment, or a recent history of such imprisonment. Execution or other political murders and brutality may be common. Unlimited detention, with or without trial, for political views is accepted...
4 - The practices of (Level 3) are expanded to larger numbers. Murders, disappearances are a common part of life...In spite of its generality, on this level terror affects primarily those who interest themselves in politics or ideas.
5 - The terrors of (Level 4) have been expanded to the whole population...The leaders of these societies place no limits on the means or thoroughness with which they pursue personal or ideological goals.
The codings are based on country reports compiled by Amnesty International
and the US State Department. The two resulting measures correlate with a factor of
.791 over the 1976-2003 period. Poe et al. (2001) report that State Department reports
have slightly favored US allies and discriminated against leftist foes in the 1970s and
early 1980s, but that these biases have mostly disappeared since. Following Poe and
36
Tate (1994), I use the Amnesty International measure, replacing missing values with
the State Department scores, where possible.22
Genocides and politicides are the most egregious forms of human rights
violations.23 I account for this fact by setting the PTS to its highest value of 5 for every
year during which genocide or politicide occurred in that country.24 By combining the
dichotomous variable for genocides/politicides with the PTS, I avoid potential problems
of collinearity between the two.25 Harff (2003, see also Harff and Gurr 1988) provides a
list of genocides and politicides, based on data collected for the State Failure project
(Goldstone et al. 2000), which have been used in empirical studies on genocides and
politicides (e.g., Krain 1997) and their consequences (Moore and Shellman 2004).
Civil wars tend to increase the level of state repression (Poe and Tate 1994;
Krain 1997; Poe et al. 1999; Zanger 2000; Harff 2003). Considering the time period for
which both civil war lists and human rights data are available, 1977-1999, and
Sambanis’ (2004) civil war data, the mean (modified)26 PTS score during 656 civil war
years is 3.94, compared to a mean of 2.20 for the 2,954 domestically more peaceful
22 Considering Amnesty scores only could introduce a sample bias problem, as Amnesty International, according to their mission, tended to over-report countries with high levels of human rights abuses in earlier years. Today, both sources cover the majority of countries in the world (Poe and Tate 1994; Poe et al. 2001). 23 Genocides are incidents of state-sponsored mass murder in which perpetrators define their victims “primarily in terms of their communal characteristics. In politicides, in contrast, groups are defined primarily in terms of their [perceived] political opposition to the regime and dominant groups” (Harff 2003, 58). 24 In total, ninety country-years were changed from a lower score to ‘5,’ including one (odd) ‘1,’ 23 country-years previously coded ‘3,’ and 64 country-years previously at ‘4.’ 79 country-years with genocide or politicide already coded ‘5’ remained unchanged. 25 Krain (1997, 331, FN1) states that the main difference between genocide/politicide and other state-sanctioned human rights abuses “is one of intentionality. The purpose behind state-sponsored mass murder […] is to eliminate an entire group.” If the main difference is one of intention and not necessarily one of behavior, a combination of the two measures seems justified. 26 From here on, PTS refers to the modified PTS, where all genocides and politicides are coded as ‘5.’
37
country years and an overall average of 2.49. In methodological terms, this could pose
problems of endogeneity.
I employ three strategies in order to counter these problems. First, I delete all
civil war years from the analyses (save the first year which is, of course, included).
Thus, cases in which an ongoing civil war led to an increase in human rights violations
while a second war happened to break out do not affect the results. The level of
analysis is the country, not the civil war. This strategy might also be helpful for the
interpretation of some other variables that could change during an ongoing civil war,
like political instability. If a first ongoing civil war causes political instability and a
second civil war erupts in the same country, a statistical relationship between instability
and civil war onset might be spurious.
Second, as mentioned above, I test my models across six different civil war lists.
Regan’s (1996) list, for example, uses a low threshold of 200 battle deaths, accounting
for what other lists would regard as lower-level violence. In general, the differences
between the start dates (Sambanis 2004, 831) should somewhat alleviate problems of
endogeneity. Even if some civil war lists code the starting date ‘too late’ for some
conflicts, others will code it correctly, guarding against an endogenous effect of civil war
on government repression.
Third, in further alternative tests, I use a different measure for human rights
violations. In an alternative set of models, I include the measure for political
imprisonment only, and exclude the variables for torture, ‘disappearances,’ and extra-
judicial killings. A repressive strategy of political imprisonment should be less closely
38
related to the lower-level violence preceding a civil war. McCormick and Mitchell (1997)
have argued that repression is a multidimensional concept that cannot easily be
captured in a single variable. They propose to disaggregate abuses of personal integrity
rights into at least the two dimensions of political imprisonment and torture/killing.
Cingranelli and Richards (1999), on the other hand, find that state repression can not
only be interpreted as a unidimensional concept; violations, further, occur in a
sequential ordering: Political imprisonment and torture occur more often and at an
earlier stage than ‘disappearances’ and killings. Cingranelli and Richards (2004) have
created a dataset that separately codes for political imprisonment, torture,
‘disappearances,’ and extra-judicial killings. Each of these variables is measured on a 9-
point scale, where, in contrast to PTS, higher scores indicate fewer human rights
abuses. In the robustness tests below, I use their measure of political imprisonment
only.
In order to avoid any conceptual overlap between human rights behavior and
democracy, I use the institutional Polity2 measure. This 21-point measure is created by
subtracting the 11-point autocracy scale from the 11-point democracy scale, resulting in
a scale that ranges from -10 (most autocratic) to +10 (most democratic). The
“institutionalized democracy” scale is composed of four additive dimensions:
competitiveness and openness of executive recruitment, constraint on the chief
executive, and competitiveness of political participation. The “institutionalized
autocracy” scale is composed of the same four dimensions, plus a dimension measuring
the regulation of participation. In the original 11-point democracy and autocracy scales,
39
years of foreign occupation are coded as interruption periods (-66), years of “complete
collapse of the central political authority,” mostly during civil war, are coded as
interregnum periods (-77), and the years during which a new polity is established are
coded as transition periods (-88). The combined Polity2 measure, however, codes only
interruption periods as missing; for interregnum periods, a ‘0’ is coded, and for
transition years, the scores are imputed by simple interpolation (Marshall and Jaggers
2002). Countries with scores from -5 to +5 are dichotomously coded as anocracies. In
order to facilitate the interpretation of the model, the anocracy variable is reversed so
that ‘0’ denotes anocracy and ‘1’ denotes no anocracy.27 As I do not necessarily expect
the effect of democracy to be consistent across anocracies and non-anocracies, I also
include an interaction term between the two variables. I employ Fearon and Laitin’s
(2003a) measure of political instability which is dichotomously coded ‘1’ if there has
been a change of at least three points on the Polity2 scale in any of the three preceding
years, and ‘0’ otherwise.
iii) Other Independent Variables
The level of economic development is approximated by GDP per capita. Although
data is available from the World Bank’s World Development Indicators, there are many
missing values, especially in the cases of interest: weak, repressive countries. Using
World Bank data, even if additional sources like the Penn World Tables are consulted,
many country years during which a civil war began would drop from the analysis. 27 The Polity2 variable is centered on ‘0,’ implying that such a country is an anocracy. The size and interpretation of coefficients and odds-ratios are not affected by the decision to reverse the anocracy variable.
40
Therefore, I use Fearon and Laitin’s data on GDP per capita. They use data from the
Penn World Tables28 and from the World Bank, but estimate missing values using data
on energy consumption from the COW project (Fearon and Laitin 2003b).
The World Development Indicators also provide data on population size. This
variable is logged (skewness = 8.08). Countries are coded as newly independent states
in their first two years after independence, similar to Fearon and Laitin’s (2003a)
measure. To operationalize military regimes, countries are dichotomously coded ‘1’ as
long as a regime led by a military person as the chief executive came to power by
means of a military coup (Madani 1992). The data are adopted from Poe et al. (2005)
who updated it through 2003. Countries whose export revenues from oil sales exceed
one third of their total export revenues are dichotomously coded as oil exporters.
Countries are coded in five-year intervals, based on World Bank data. Rough terrain is
measured as the percentage of mountainous terrain, based on the codings of
geographer A. J. Gerard. Missing values are estimated using the difference between the
highest and lowest point of elevation (see Fearon and Laitin 2003b). Measures for
ethnic and religious fractionalization denote the probability that two randomly chosen
individuals in a country are from two different ethnic or religious groups, respectively.
The measures are based on the 1964 Soviet Atlas Narodov Mira. The measures for oil
exports, mountainous terrain, and ethnic and religious fractionalization are directly
adopted from Fearon and Laitin’s dataset (2003a, 2003b).
28 http://pwt.econ.upenn.edu/, accessed October 2004.
41
These independent variables do not show strong signs of problems of
multicollinearity. Except, of course, for the interaction term between democracy and
anocracy and democracy by itself, no two variables are correlated higher than |.5| and
for only four pairs of variables is the correlation coefficient higher than |.4|: Polity2 and
government repression (-.45), and the log of GDP per capita with Polity2 (.48), with the
presence of a military regime (-.43), and with ethnic fractionalization (-.48). All non-
dichotomous independent variables, except for state repression and democracy,29 are
centered on their means, to facilitate interpretation of the statistical outputs. I subtract
‘1’ from the PTS so that the constant in a logistic regression model now represents the
logarithm of the probability of civil war onset in any given year for a country with
average levels of economic development, population size, mountainous surface, and
ethnic and religious fractionalization. This ‘average country’ has almost no repression
(PTS=1), is an anocracy (since Polity2 = 0), has experienced no political instability in
the past three years, is not led by a military government, and is not an oil exporter (all
dummies are set to ‘0’). The descriptive statistics for all independent variables in their
raw form, i.e. before they are lagged, logged or centered on their mean, are
summarized in Table A2 in the appendix.
B. Results
I first present the results with Sambanis’ (2004) civil war list as the dependent
variable in detail and then, more shortly, the robustness tests across model
29 The Polity2 scale is not centered because the mean (-.031) is very close to zero.
42
specifications with alternative independent and dependent variables. The results from a
logistic regression with Sambanis’ (2004) civil war list are summarized in Table 1.30 The
table gives the coefficients and odds-ratios with their respective standard errors, and
the p-values. These results are compared to the Null-Model without the repression
variable. The coefficients of a logistic regression cannot be easily interpreted in
substantive terms, but the odds-ratios indicate the factor by which the odds of civil war
onset increase with a one-unit increase in the independent variable. Odds, in turn, are
the chances of seeing a civil war divided by the chances of not seeing one. First,
though, I only interpret signs and levels of significance.
In the Null-Model in Table 1 state terror is omitted from the list of variables. As
the results for other independent variables only change marginally, I only discuss the
results from Model 1 in detail. This congruence between the Null-Model and Model 1
indicates that repression adds considerably to the explanation of civil war onset without
altering the effects of other variables, contrary to Hypothesis 2a, not even that of
democracy.
Of the state weakness variables, the level of economic development is
statistically significant. This seems to be one of the most robust findings in the
empirically oriented civil war literature. Surprisingly, population size does not even come
close to statistical significance and the coefficient’s sign is not in the expected direction,
while it was one of the most robust variables in Sambanis (2004) and Fearon and Laitin
30 To test for robustness, I alternatively estimate this and all following models with Probit maximum likelihood estimations. In further tests, I replace the modified PTS that is based on Amnesty International reports by the one that is based on US State Department country reports (replacing missing values with the Amnesty scores, if possible). In both cases, the results are similar to the ones reported here. (The Probit and State Department models are not reported.)
43
(2003a) and significant in at least some of Collier and Hoeffler’s (2001) models. New
states, on the other hand, are much more likely to experience a civil war. In fact, this
variable was dropped from the analysis since it predicted failure perfectly. Sambanis
(2004, 837), despite his criticizing the theoretical expectations related to new states,
reports similar findings.
Next, civil wars are significantly more likely to occur in countries governed by
military regimes. This finding adds to existing theory on the causes of civil wars, as it
was not tested in any of the three studies cited above. Political instability, contrary to
earlier findings (Fearon and Laitin 2003a; Sambanis 2004), was not significant. Contrary
to my expectations, democracy positively impacted the probability of civil war onset.31
However, democracy’s positive impact is only statistically significant within the group of
anocracies, where Polity2’s coefficient is .249. For non-anocracies, the coefficient
reduces to almost zero at .04, and is no longer statistically significant, meaning that
fully consolidated democracies and fully established authoritarian regimes are equally
likely to experience a civil war, ceteris paribus.32 Taken together, both democracies and
autocracies are still less likely to experience civil war than anocracies, as indicated by
the negative and significant coefficient. Taken together, these findings underline the
importance of modeling specific aspects of a country’s political system, rather than just
including the overall level of democracy. Yet, they clearly deserve increased
31 The level of democracy also had a positive coefficient in Fearon and Laitin’s (2003a) Model 1 but was not statistically significant. 32 On the interpretation of interaction terms, see Braumoeller (2004). That the coefficient is not significant for non-anocracies can be seen from an alternative model, where the anocracy variable is not reversed, i.e. anocracies are coded ‘1’ and non-anocracies are coded ‘0’ (this model is not reported). Here the p-value of Polity2 is .211.
44
examination in future studies, especially since all other variables related to a country’s
political system were signed in the expected direction. Wantchekon (2004) describes
how democracy can arise from civil war, but how does a higher level of democracy
(within anocracies) lead to civil war?
Figure 2 (and Table A3 in the appendix) describe the bivariate relationship
between the level of democracy and civil war onset in greater detail, not controlling for
other factors. As can be seen from Figure 2, fully established democracies are very
unlikely to experience a civil war. In absolute terms, a relatively high level of new civil
wars seem to break out in almost fully established authoritarian regimes (at Polity2 = -7
and -6). In relative terms, however, anocracies, and especially more democratic
anocracies (between 0 and +5) experience the highest level of civil war onsets.33 This
finding deserves increased scholarly attention in the future. One explanation might be
that half-hearted democratization processes open accesses to the political system for a
number of groups without effectively sharing political power with these groups. At the
same time, state leaders may feel international and domestic pressures to democratize,
but are not willing to give up power. If they pretend to give in to these pressures and
hold meaningless elections, this might be a signal of state weakness. A full
democratization, on the other hand, firmly includes contending groups in the polity,
making violent rebellion an unnecessary choice. Democracy, however, may have a
different effect on other aspects of civil war. Lacina (2005) finds democracy to be one
of the strongest factors in reducing the number of battle deaths in a civil war, once 33 If Polity2 is replaced by either the Freedom House political rights or civil liberties scale, both have negative signs and respect for political rights significantly decrease the probability of civil war onset, while civil liberties do not seem to have an impact (results not reported here).
45
started. Elbadawi and Sambanis (2002) find that democracy reduces civil war
prevalence. The link between democracy and civil war seems to be more complicated
than the democratic peace in international politics.34 In From Voting to Violence, Snyder
(2000) describes how, under certain circumstances, democratizing and pseudo-
democratic countries may experience nationalist or ethnic conflict.
Oil exporting states face a higher risk of seeing a civil war. As argued above, oil
revenues present an incentive for rebels to initiate an insurgency and this finding
supports Collier and Hoeffler’s (2001) thesis of greed-based rebellions. When the spoils
of a civil war are high, potential rebel elites will react upon opportunities to realize these
spoils. The variables for ethnic and religious fractionalization are not significant, neither
is each of these variables, when the other one is excluded from the model.35 Similarly,
the percentage of mountains of a country’s surface does not seem to be related to the
probability of civil war onset, according to this model. Likelihood-ratio tests reveal that
any of the statistically insignificant variables do indeed not offer additional explanatory
power for civil war onset.
The repression-rebellion link, finally, receives major support from the findings of
this model: Past repression exerts a highly significant impact on the probability of a civil
war to break out. In fact, the coefficient of government repression is the most
significant in statistical terms. The odds-ratio allows a first glimpse at the substantial
impact of this variable: For each 1-point increase on the Political Terror Scale, the odds
34 But even with regard to interstate war, Mansfield and Snyder (2002) report findings that incomplete transitions toward democracy increase the probability of war. 35 The results for these alternative model specifications, with ethnic fractionalization only and religious fractionalization only, are not reported.
46
of civil war onset increase by a factor of 2.039. For a change from the lowest to the
highest level of repression, the odds of civil war onset would increase by a factor of
17.28. Figure 3 (and Table A4 in the appendix) describe the bivariate relationship
between the modified Political Terror Scale and civil war onset.
As can be seen from Figure 3, every increase in repression seems to increase the
probability of civil war onset. However, I also tested for the hypothesized, and often
empirically supported, curvilinear impact of repression on domestic political violence
discussed in previous research (Gurr 1970; Tilly 1978; Muller 1985; Muller and Seligson
1987; Muller and Weede 1990; Boswell and Dixon 1990; Schock 1996).36 These tests do
not show statistically significant results, with or without PTS in the model (results not
reported). Any increase in the level of repression, therefore, seems to increase the
probability of civil war onset, not just up to intermediate levels of repression. Next, I
turn to an interpretation of the substantial effects of the independent variables.
C. Substantial Interpretation and Simulations
The substantial interpretation of coefficients in logistic regression is not as
straightforward as the interpretation of their signs and levels of statistical significance.
For this reason, I have generated a table with the simulated substantial effects of the
significant variables, holding all other variables at their means and setting dichotomous
variables and democracy to ‘0.’ In Table 2, the effect of changes on the statistically
significant variables is simulated. As a point of reference, I calculate the probability of
36 By adding a variable that assumed a value of 0 if PTS = 3, 1 if PTS = 2 or 4, and 2 if PTS = 1 or 5.
47
an ‘average country’ to experience the outbreak of a civil war. For this hypothetical
country, all dichotomous variables and the Polity2 democracy measure are set to ‘0,’
PTS is set to ‘1,’ and the other independent variables are at their respective means. The
expected probability of civil war onset for the ‘average country’ would be .51% in each
year. Clearly, this is a very low probability, even over ten years, the accumulated
probability increases to only 5%. Table 2 denotes the change in the significant
variables, the expected probabilities of civil war onset, and the ratio of these
probabilities over that of the ‘average country.’ In general, one should expect relatively
low probabilities of civil war onset, as it only occurs in about 2% of all country years.
Increases of one unit on any of the independent variables have different effects
depending at which level they occur. For example, a change from 1 to 2 points on PTS
has a different impact than a change from 2 to 3 points; and both effects vary with the
level of other independent variables.
In addition, Table 2 gives the expected probabilities over a time span of five and
ten years, given that the independent variables do not change. For the ‘average
country,’ for example, given that it stays at the means of all independent variables, a
probability of .51% per year accumulates to about 2.5% over five years and to about
5% over a decade. If, on the other hand, such a country experienced a democratization
process, for example, or an economic crisis, the predicted probabilities would change.
48
i) The Substantial Impacts of the Independent Variables
First, an increase of 1 point on the Political Terror Scale would result in an
expected probability of 1.04%, about 2.03 times higher than the probability of the
‘average country.’ An increase by 4 points from the lowest to the highest level of
repression would lead to a probability of 8.21%, almost 16 times higher than for a
country without repression, at a level of ‘1’ on the Political Terror Scale. Over a decade,
an accumulated probability of almost 60% would be expected. In fact, government
repression is the variable that leads to the single largest increase in the probability of
civil war onset when set to its maximum.37 Clearly, repressive governments are much
more likely to be faced with a civil war; repression, it seems, is provocation much more
than deterrence.
Second, the effect of democracy, although not in the expected direction, is only
significant for anocracies. Within the group of anocracies, a change towards more
democracy exerts a sizable impact. The most authoritarian anocracy (at Polity2 = -5)
faces a risk of civil war risk of only 15%, 29% that of the ‘average country.’ For the
most democratic anocracy (Polity2 = +5), on the other hand, the expected probability
of civil war onset is 1.77%, 3.44 times as high as the ‘average country.’ In non-
anocracies, regardless of the level of democracy, the probability reduces to .24%, less
than half that of the average country. While democracy does not seem to play a role for
non-anocracies, the finding that the level of democracy increases the likelihood of civil
war onset within the group of anocracies clearly deserves additional scrutiny.
37 A change from the highest to the lowest level of economic development would increase the expected probability of civil war onset by a factor of almost 15.
49
Third, military regimes greatly increase the probability of civil war onset. The
presence of a military regime increases the likelihood of war onset by a factor 2.25 to
1.16%, holding all other variables, including the level of democracy, constant.
Fourth, the level of economic development had the second largest impact on the
likelihood of civil war onset. An increase from the mean of $2,59838 (constant 1985 US
dollar) to the maximum of $28,901 (United Arab Emirates in 1980) leads to a decrease
in the expected probability to .13%, only 25% that of the ‘average country.’ On the
contrary, a move down to the minimum of $215 (Democratic Republic of the Congo in
1998) leads to an increase to 1.94%, 3.8 times more than the ‘average country.’ Over a
time period of ten years, a country with a maximum level of economic development
faces a probability of war onset of only 1.3%, whereas it is almost 18% for a country at
the minimum. Comparing the two, countries with a minimal level of economic
development were more than 15 times as likely to experience the outbreak of a civil
war as countries with a maximal level of economic development. While the importance
of economic development for civil war onset is one of the most robust findings in the
literature, its practical importance may be relatively low. At least in the short term,
economic development changes rather slowly, whereas repression can be considerably
reduced even over the period of a year or two.
Fifth, oil exporters face a much higher probability of civil war onset, at 1.40%,
2.72 times higher than the ‘average country’ without oil exports. In this model, oil
38 This is the mean of the logged and lagged variable; it is different from the one reported in Table A1 in the appendix.
50
exports represent the dichotomous variable with the strongest impact (not regarding
the new states variable).
ii) Hypothetical Cases: The Four Subgames
How do these effects combine in hypothetical cases? More specifically, what
probabilities of civil war onset can be expected in the ideal types of a strong and a
weak state, with and without repression (i.e. in the four subgames)? The ideal type of a
strong state, in this simulation, refers to a fully consolidated democracy or a fully
established authoritarian (not an anocracy), but not a military regime with a maximum
level of economic development, but no or low-level oil exports. The ideal type of a weak
state refers to an anocracy at Polity2 = 5, governed by a military regime, with oil
exports and a minimum level of economic development. In this hypothetical case, the
opportunity costs of rebelling are low and the potential spoils are high. For both cases,
population, the percentage of mountainous terrain, ethnic and religious fractionalization
are held at their respective means, and both are politically stable, since these variables
were not significant in the Model 1. To represent the four subgames in Figure 1, both
weak and strong states without (PTS = 1) and with high levels of repression (PTS = 5)
are simulated. Strong states with no (or almost no) state-sponsored human rights
abuses, as expected, are the least likely to experience a civil war; the probability
reduces to .06%, only 12% of the probability of the ‘average state.’ Strong states with
high levels of repression are 17.67 times more likely to face a civil war, with a
probability of civil war onset of 1.06% (2.06 times that of the ‘average country’). The
51
effect of state weakness, however, (combined with a low n, a high s, and no
repression) is much more dramatic, increasing the likelihood to 29.96% (more than 58
times that of the ‘average country’ and 500 times that of a strong state with no
repression). Clearly, while repression has the single most substantial effect on civil war
onset, the combination of different aspects of state weakness, large potential spoils,
and low opportunity costs, outweighs the impact of the single repression variable. The
game suggests that state weakness is a necessary condition for a civil war to start,
while state weakness and repression combined are sufficient. Accordingly, repression
still increases the probability of civil war onset, even in weak states. Combining state
weakness and repression, the expected probability of civil war onset skyrockets to 88%,
almost three times higher than without repression, 170 times higher than the ‘average
country,’ and 1,470 times higher than in a strong state with no government
repression.39 From these simulated probabilities, then, one can conclude that state
weakness and repression come quite close to being sufficient conditions.
These simulated probabilities strongly support the theorized combined effect of
state weakness and repression – flanked by low opportunity costs and large potential
spoils – on the probability of civil war onset. Even in a perfectly weak state, the
probability of civil war onset is only about 30%, whereas it is 90% in a weak and
repressive state. How do these simulations compare to actual predicted probabilities of
civil war onset in the sample? The lowest probability is calculated for the United Arab
Emirates in 1983, with .04% equaling that of the ideal type of a strong state with no 39 One might argue that this effect should empirically be tested by using interaction terms. Yet, interaction terms between repression and economic development and, alternatively, repression and democracy, were not significant (results not reported).
52
repression. Indeed, the UAE come close to this ideal type on most independent
variables. The highest probability, on the other hand, is predicted for Uganda in 1981 at
57%. This country comes close to the ideal type of a weak state with widespread
human rights violations (PTS = 5, Polity2 = 3); a civil war broke out that same year.
D. Robustness Tests
Sambanis (2004) makes a strong argument for testing the robustness of civil war
models over different civil war coding schemes. Using the exact same set of
independent variables, I replace the dependent variable (Sambanis’ 2004 list of civil
wars) with five alternatives: the COW list of civil wars (Sarkees and Singer 2001), the
Uppsala Armed Conflict Dataset codings (Gleditsch et al. 2001), Fearon and Laitin’s
(2003a), Regan’s (1996), and Doyle and Sambanis’ (2000, extended) measures. All the
data are adopted from Sambanis’ (2004) dataset. As discussed above, I chose these
alternatives because they represent the most widely used datasets, the most recent and
most widely quoted studies on civil war, and different coding rules to identify instances
of civil war.
As can be seen from Table 3, government repression is the most robust variable
throughout Models 2 to 6.40 With the COW list, the p-value increases to 4.4%, and with
Regan’s list, it increases to 2.5%, while it stays below .001% in all other cases. The
variables for democracy, anocracy, military regimes, and oil exports are much less
robust, despite significant results in Model 1. Democracy, within anocracies, is 40 I also added the modified PTS scores (based on Amnesty International and, alternatively, US State Department country reports) to Fearon and Laitin’s (2003a) Model 1, and found them highly significant, with and without ongoing civil wars included. The results are not reported here.
53
significant in three out of five models, increasing the probability of civil war onset.
Anocracy, however, is never statistically significant. Similarly to democracy, GDP per
capita is only significant in three models, while Sambanis (2004) found it to be one of
the most robust variables in predicting civil war onset. The results for new states are
much more robust; the variable drops out of each model. In contrast to the results from
Model 1, population size and the percentage of mountainous terrain seems to exert
significant impact on the probability of civil war onset in one of the models each.
Finally, the results for political instability, ethnic and religious fractionalization are very
robust, showing no effect on civil war onset in any of the six models, although political
instability comes close to significance in Model 2.
To summarize, repression, relatively democratic anocracies, low economic
development, and recent independence seem to be important factors that strongly
increase the probability of civil war onset, whereas political instability and ethnic and
religious fractionalization do not seem to matter much. The results for military regimes,
oil exports, population size, and the percentage of mountainous terrain, on the other
hand, are more ambiguous, with two of the coefficients changing signs and varying
levels of significance.
As an alternative test of the robustness of the impact of repression on civil war
onset, I replace the general measure of human rights violations with a variable that
only takes into account political imprisonment. Cingranelli and Richards (2004) provide
detailed data on all four dimensions of the Political Terror Scale, of which I use the
measure for political imprisonment. This variable, as the three others, is closely related
54
to both Cingranelli and Richards’ aggregate measure of human rights abuses (.77) and
the modified Political Terror Scale (-.61). In contrast to the Political Terror Scale, higher
measures on all of Cingranelli and Richards’ measures denote fewer abuses.
As can be seen from Table 4, the impact of government repression on civil war
onset is still significant with this alternative measure. In this table, only the results for
Sambanis’ (2004) list of civil wars are summarized; further robustness tests across
alternative civil war lists are deferred to the appendix. In Model 7 (Table 4), none of the
effects of other independent variables change considerably. In Models 10 through 12
(Table A5 in the appendix), in contrast, political imprisonment is not significantly
correlated to civil war onset. This is the only caveat found to the robustness of the main
findings. Possibly, although correlated, political imprisonment alone is not harsh enough
a condition to ignite people to rebel. In terms of the game-theoretic model, political
imprisonment might not increase the costs of not joining a rebellion (r) to a level
sufficiently high as to making participation a rational choice. On the other hand, in three
out of six models, political imprisonment does significantly increase the likelihood of a
new civil war. Overall, these robustness tests still seem to support the repression-
rebellion link.
To summarize, the effect of state repression on civil war onset is found to be
significant across different measures of repression and different dependent variables.
This finding qualifies earlier findings on the importance of opportunity and willingness,
and greed and grievance in an important way. While weak states provide an
opportunity for greedy rebel elites to realize war spoils, nonelites join when the costs of
55
not joining become too high. These individually perceived grievances make them willing
to participate.
56
C O N C L U S I O N
State-sponsored repression is an important factor in increasing the probability of
civil war onset. This finding is robust across alternative civil war lists and different
measures of repression. Further, as can be seen from the simulated effects of state
weakness and repression, it is when these factors operate simultaneously that the risk
of a new civil war reaches high levels. This finding provides support for the argument
that even in weak states governments have a choice between confrontational and
accommodative strategies and can significantly reduce the probability of a civil war to
break out. The game-theoretical model of civil war onset receives similar support from
these findings and simulations.
Rather than restating the results in detail, I conclude by discussing possible
implications and options for further research. First, the findings in this study suggest
that repression leads to civil war, but it is not clear, how exactly this happens. Future
research should examine this process more closely, maybe rather in comparative case
studies than with a large-n research design. The strategic interplay between emerging
rebel movements and the government deserves special attention in this context.
Further, it should be examined whether and when rebel organizations emerge from
formerly nonviolent opposition movements or whether these are separate groups in
most cases. Future research could also investigate the role of human rights abuses
committed by the rebels. If the rebels have to rely on popular support, why do they
sometimes repress the population, as in Uganda, the Democratic Republic of the Congo,
or elsewhere?
57
Second, the link between democracy and civil war onset, but also civil war
duration, intensity, and outcome should be examined more closely. Clearly, the
relationship between democracy and conflict is not as straightforward in the domestic
sphere as it is in international politics. If a democratic domestic peace does exist (Krain
and Myers 1997; Hegre et al. 2001), it embraces more than just the level of institutional
democracy. Also, other factors could be included in multivariate models, for instance,
measures of the international situation (e.g., Boswell and Dixon 1990).
Third, rather than looking at repression alone, future studies could focus on how
governments accommodate opposition demands, and what effects accommodation has
on opposition group strategies. Chong (1991), for example, argues that accommodation
could have the counterintuitive effect of encouraging the opposition, leading to more
extreme demands.
Fourth, repression drives people into rebellion, but it can also lead to forced
migration (Mason and Krane 1989; Azam and Hoeffler 2002). Empirical research on the
causes of refugee movements has consistently found state terror, including politicide
and genocide, to be one of the strongest predictors (Schmeidl 1997; Davenport et al.
2003; Moore and Shellman 2004). Future research should look at the conditions
favoring ‘voice’ over ‘exit,’ or those leading to both responses.
Fifth, the practical relevance of these and related theoretical findings should be
explored. Civil wars are more deadly than interstate wars (Gurr 1970, 3), occur more
frequently (Fearon and Laitin 2003), produce refugee movements (Schmeidl 1997;
Davenport et al. 2003; Moore and Shellman 2004) and a further increase in state
58
repression (Poe and Tate 1994; Poe et al. 1999; Zanger 2000), open the way for
politicide and genocide (Krain 1997; Harff 2003), lead to long-term civilian suffering
(Ghobarah et al. 2003), and provide the environment for organized crime and terrorism
(UN 2004). Specifically, the possibilities of creating an early warning or risk assessment
device for civil war onset should be explored more carefully. Early warning and risk
assessment models have been constructed to forecast a number of events in the social
sciences (see Davies and Gurr 1998 and Carment 2003 for an overview). Looking at the
predicted probabilities of civil war onset (the p-hats), only three civil wars broke out in
countries with a p-hat below the median of .86%, while 47 (a rate of 3.84% over all
country-years) occurred above the median. In the upper 25% of the p-hats (above a
probability of 2.24%), the rate of civil war onsets increases to 5.88%; in the upper 10%
(above a probability of 4.44%) the rate of onsets increases to 10.20%; in the upper 5%
(above a probability of 6.98%) the rate of onsets increases to 13.82%, and in the upper
1% (above a probability of 19.44%) the rate of onsets increases to 25%. This means
that the risk to experience a civil war of the upper 1% is almost 14 times higher than
that of the remaining 99% of the country-years. This indicates a potential for risk
assessment devices, but it remains to be examined whether out-of-sample assessments
are similarly possible.
The results presented in this study modify earlier findings on civil war onset in an
interesting way. They may yield some practical implications and pose questions for
future research. Repression, especially in weak states, paves the way for civil war,
leaving governments with the choice of repressing and risking a war or accommodating
59
opposition demands. This study shows that the choice for repression, despicable on
humanitarian and moral grounds alone, may entail even more disgraceful
consequences.
60
Figure 1. The onset of civil war in an extensive game form.
repression r = 1
-d; n – r s + p; m
n – r; n – r
repression r = 1
no repression r = 0
no repression r =0
n; n
j j
j j
~j ~j
~j ~j
G
i
i i
~i
~i
C
CC
C
~i
~i
E
G
weak state w = 1
strong state w = 0
Nature
E
EE
i
1 – d; -d
n – r; n – r
s; n – r
n; n
s + p; m s; n -d; n1 – d; -d
61
Table 1. Logit regression on the probability of civil war onset. Null-Model Model 1 Sambanis (2004) Sambanis (2004) coef. p-value coef. odds ratio p-value Pesonal integrity abusesa - - 0.713*** 2.039*** 0.000 (0.161) (0.328) Democracy (Polity2)a 0.250*** 0.001 0.249*** 1.283*** 0.001 (0.075) (0.078) (0.100) No anocracya -0.833** 0.020 -0.749** 0.473** 0.037 (0.357) (0.359) (0.170) (Polity2 * No Anocracy)a -0.238*** 0.002 -0.209*** 0.811*** 0.009 (0.077) (0.081) (0.065) Political instability 0.114 0.767 -0.048 0.954 0.902 (0.385) (0.384) (0.366) Military regimea 0.937*** 0.009 0.816** 2.262** 0.026 (0.361) (0.366) (0.828) New states
dropped Dropped
GDP per capitaa, b, c -0.615*** 0.009 -0.548** 0.578** 0.020 (0.236) (0.236) (0.136) Population sizea, b, c 0.173 0.103 -0.013 0.987 0.910 (0.106) (0.118) (0.117) Oil exporting countriesa 1.069** 0.011 1.009** 2.742** 0.017 (0.419) (0.422) (1.158) % mountaineousb 0.010 0.121 0.006 1.006 0.380 (0.006) (0.007) (0.007) Ethnic fractionalizationb 0.160 0.799 0.106 1.111 0.868 (0.627) (0.635) (0.706) Religious fractionalizationb 0.375 0.617 0.820 2.269 0.306 (0.750) (0.801) (1.817) Constant -4.212 0.000 -5.264 0.000 (0.391) (0.482) N 2527 2450 Likelihood-Ratio chi² (11) 66.91 85.2 Pseudo R² 0.1362 0.1745 Note: standard errors are in parentheses * p < .1, ** p < .05, *** p < .01 a lagged by one year b centered on the mean c natural log taken
62
02
46
810
1214
1618
-10 -8 -6 -4 -2 0 2 4 6 8 10
Polity2 (lagged)
# civil war onset% civil war onset
Figure 2. Civil war onset and the (lagged) level of democracy.
02468
1012141618
1 2 3 4 5
modified PTS (lagged)
# civil war onset%civil war onset
Figure 3. Civil war onset and the (lagged and modified) PTS.
63
Table 2. Simulated effects of statistically significant variables.
change in independent variable
probability of civil war onset this year
ratio of probability / probability of 'average country'
probability of civil war onset over 5 years
probability of civil war onset over 10 years
'average-country' 0.51% 1.00 2.55% 5.03%
personal integrity abuses +1 (from 1 to 2) 1.04% 2.03 5.11% 9.97%
+4 (from 1 to 5) 8.21% 15.95 34.85% 57.56%
Democracy/Anocracy Polity2 = +5; anocracy 1.77% 3.44 8.54% 16.35%
Polity2 = -5; anocracy 0.15% 0.29 0.74% 1.47%
no anocracy 0.24% 0.47 1.21% 2.42%
No military regime military regime 1.16% 2.25 5.65% 10.99%
GDP per capita + 1 std. dev. (1985 US
$2,933) 0.29% 0.56 1.42% 2.82%
to max: 1985 US $28,901
(UAE) 0.13% 0.25 0.65% 1.30%
to min: US $215 (DR
Congo) 1.94% 3.77 9.34% 17.81%
No oil exporter oil exporter 1.40% 2.72 6.80% 13.14%
strong state, no repression 0.06% 0.12 0.31% 0.62% weak state, no repression 29.96% 58.19 83.14% 97.16% strong state, high repression 1.06% 2.06 5.18% 10.09% weak state, high repression 88.09% 171.09 100.00% 100.00% lowest predicted chance in sample United Arab Emirates, 1983 0.04% 0.09 0.22% 0.45% highest predicted chance in sample Uganda, 1981 57.21% 111.11 98.56% 99.98% Note: The 'average-country' is a country with all dummies and Polity2 set to '0', PTS at ‘1,’ and all other variables at their mean. A ‘strong state’ has an authoritarian, stable, and non-military government (Polity2 = -10, no anocracy), a high GDP per capita of $28,901 (in 1985 US constant $), is not an oil exporter, and has an average population size and average levels of mountainous terrain, ethnic and religious fractionalization. A ‘weak state’ is a stable anocracy (Polity2 = 5) that is governed by a military regime, has a low GDP per capita of $215, is an oil exporter, shows an average population size and average levels of mountainous terrain, ethnic and religious fractionalization. ‘No repression’ refers to a low level of ‘1’ on the PTS, ‘high repression’ to a level of ‘5.’
64
Table 3. Logit regression on the probability of civil war onset. Model 2 Model 3 Model 4 Model 5 Model 6
coef. p-value coef. p-value coef. p-value coef. p-value coef. p-value
Pesonal integrity abusesa 0.358 0.044 1.291 0.000 0.737 0.000 0.403 0.025 0.602 0.000
(0.177) (0.164) (0.189) (0.180) (0.165)
Democracy (Polity2)a -0.023 0.815 0.160 0.046 0.285 0.002 -0.044 0.733 0.159 0.055
(0.097) (0.081) (0.091) (0.129) (0.083)
No anocracya -0.333 0.436 -0.507 0.162 -0.559 0.187 0.048 0.925 -0.569 0.124
(0.427) (0.363) (0.423) (0.515) (0.371)
(Polity2 * No anocracy)a 0.027 0.789 -0.147 0.075 -0.254 0.007 0.077 0.555 -0.162 0.060
(0.100) (0.082) (0.095) (0.130) (0.086)
Political instability 0.693 0.106 -0.027 0.945 0.092 0.828 0.258 0.552 0.301 0.437
(0.428) (0.386) (0.423) (0.434) (0.387)
Military regimea 0.303 0.439 0.092 0.803 0.532 0.181 0.248 0.517 0.744 0.040
(0.392) (0.368) (0.397) (0.383) (0.363)
New states dropped dropped dropped dropped dropped
GDP per capitaa, b, c -0.526 0.044 -0.234 0.317 -0.742 0.006 -0.829 0.003 -0.337 0.160
(0.261) (0.234) (0.269) (0.277) (0.240)
Population sizea, b, c 0.138 0.263 -0.018 0.877 0.147 0.277 0.231 0.074 0.061 0.595
(0.123) (0.115) (0.135) (0.130) (0.115)
Oil exporting countriesa 0.663 0.157 0.146 0.737 0.462 0.369 1.047 0.018 0.460 0.297
(0.468) (0.433) (0.513) (1.047) (0.441)
% mountaineousb 0.013 0.063 -0.003 0.719 0.010 0.154 0.002 0.777 0.003 0.704
(0.007) (0.007) (0.007) (0.008) (0.007)
Ethnic fractionalizationb 0.050 0.940 0.244 0.692 -0.154 0.820 0.674 0.316 0.644 0.322
(0.664) (0.615) (0.678) (0.672) (0.651)
Religious fractionalizationb 0.545 0.557 0.634 0.459 0.181 0.843 -0.505 0.568 0.648 0.431
(0.927) (0.857) (0.917) (0.885) (0.822)
Constant -5.261 0.000 -6.682 0.000 -5.734 0.000 -5.237 0.000 -5.273 0.000
(0.548) (0.559) (0.574) (0.601) (0.473)
N 2315 2781 2444 1828 2509
Likelihood-Ratio chi² (11) 44.41 124.39 74.84 46.83 60.84
Pseudo R² 0.117 0.237 0.187 0.1193 0.1325 Note: standard errors are in parentheses; coefficients with p < .1 are in bold letters a lagged by one year b centered on the mean c natural log taken
65
Table 4. Logit regression on the probability of civil war onset. Overall repression replaced by political imprisonment. Model 7 Sambanis (2004) coef. odds ratio p-value (Absence of) Political Imprisonmenta -0.606** 0.546** 0.043 (0.300) (0.164) Democracy (Polity2)a 0.346*** 1.413*** 0.000 (0.095) (0.134) No anocracya -0.804* 0.448* 0.057 (0.422) (0.189) (Polity2 * No anocracy)a -0.252*** 0.777*** 0.007 (0.093) (0.072) Political instability -0.392 0.675 0.408 (0.474) (0.320) Military regimea 1.299*** 3.667*** 0.007 (0.479) (1.757) New states
Dropped
GDP per capitaa, b, c -0.897*** 0.408*** 0.002 (0.294) (0.120) Population sizea, b, c 0.186 1.204 0.198 (0.144) (0.174) Oil exporting countriesa 1.183** 3.263** 0.024 (0.523) (1.708) % mountaineousb 0.007 1.007 0.398 (0.008) (0.008) Ethnic fractionalizationb -0.683 0.505 0.364 (0.752) (0.380) Religious fractionalizationb 1.286 3.620 0.177 (0.952) (3.447) Constant -4.086 0.000 (0.579) N 1925 Likelihood-Ratio chi² (11) 61.21 Pseudo R² 0.1834 Note: standard errors are in parentheses * p < .1, ** p < .05, *** p < .01 a lagged by one year b centered on the mean c natural log taken
66
A P P E N D I X
67
Table A2. Descriptive statistics of the independent variables.
Mean Std. Dev. Minimum Maximum Pesonal integrity abuses 2.496 1.220 1 5 Democracy (Polity2) -0.031 7.548 -10 10 No anocracy 0.855 0.352 0 1 Political instability 0.152 0.359 0 1 Military regime 0.247 0.432 0 1 New states 0.022 0.148 0 1 GDP per capita (in 1000) 4.428 4.649 0.196 31.969 Population size (in mio.) 30.329 111.380 0.017 1288.400 Oil exporting countries 0.159 0.366 0 1 % mountaineous 17.478 21.451 0 94.300 Ethnic fractionalization 0.409 0.284 0.001 0.925 Religious fractionalization 0.381 0.218 0 0.783 Note: These are the descriptive statistics for the raw data and not for the logged, lagged, or centered data. Not all the data are included in each analysis, due to missing data on other variables.
Table A1. Descriptive statistics of the civil war lists.
Years
covered
# of country-
yrs
Number of civil wars
% of civil war onsets
Number of civil war-years
% of civil war-years
Sambanis (2004) 1976-1999 3146 62 1.97% 673 21.39% COW 2000 1976-1997 2822 46 1.63% 352 12.47% Gleditsch et al. (2001) 1976-1999 3146 62 1.97% 279 8.87% Fearon & Laitin (2003) 1976-1999 3146 49 1.56% 645 20.50% Regan (1996) 1976-1994 2128 48 2.26% 571 26.83% Doyle & Sambanis (2000, extended)
1976-1999 3146 55 1.75% 604 19.20%
Note: Some observations may drop out in the analyses, due to missing data on other variables.
68
Table A3. Civil war onset and the (lagged) level of democracy.
Polity2
(lagged) No civil war
starts Civil war starts Total # of
country-years %
onset -10 114 1 115 0.88 -9 198 3 201 1.52 -8 157 3 160 1.91 -7 528 11 539 2.08 -6 142 5 147 3.52 -5 73 1 74 1.37 -4 42 1 43 2.38 -3 46 2 48 4.35 -2 63 4 67 6.35 -1 35 1 36 2.86 0 26 4 30 15.38 1 26 0 26 0 2 16 0 16 0 3 12 2 14 16.67 4 35 3 38 8.57 5 54 6 60 11.11 6 113 0 113 0 7 85 3 88 3.53 8 156 3 159 1.92 9 159 1 160 0.63 10 554 1 555 0.18
Note: The calculations are based on country-years.
Table A4. Civil war onset and (lagged) repression. PTS (lagged) No civil war starts Civil war starts Total % onset
1 592 0 592 0.00% 2 851 14 865 1.65% 3 692 16 708 2.31% 4 215 13 228 6.05% 5 50 7 57 14.00%
69
Table A5. Logit regression on the probability of civil war onset. Overall repression replaced by political imprisonment and torture.
Model 8 Model 9 Model 10 Model 11 Model 12 COW 2000 Gleditsch et al. (2001) Fearon & Laitin (2003) Regan (1996) Doyle & Samb. (2000, ext.) Coef. p-value Coef. p-value Coef. p-value Coef. p-value Coef. p-value (Absence of) -0.750 0.047 -1.248 0.000 -0.103 0.752 -0.008 0.981 -0.448 0.139 Political Imprisonment 0.378 0.320 0.326 0.324 0.303 Democracy (Polity2)a 0.028 0.826 0.320 0.000 0.350 0.001 -0.009 0.954 0.283 0.005 0.127 0.091 0.102 0.150 0.100 No anocracya -0.359 0.513 -0.665 0.089 -0.982 0.048 0.022 0.972 -0.590 0.182 0.550 0.391 0.497 0.624 0.442 (Polity2 * No anocracy)a 0.018 0.887 -0.251 0.006 -0.291 0.005 0.045 0.762 -0.256 0.011 0.128 0.090 0.103 0.149 0.100 Political instability -0.008 0.990 -0.535 0.266 -0.283 0.598 0.056 0.918 0.021 0.966 0.587 0.480 0.537 0.549 0.492 Military regimea 0.740 0.158 0.939 0.030 1.272 0.015 0.688 0.202 1.599 0.001 0.523 0.433 0.525 0.539 0.489 New states
dropped dropped dropped dropped dropped
GDP per capitaa, b, c -0.575 0.094 -0.500 0.067 -0.819 0.014 -0.949 0.007 -0.361 0.226 0.343 0.273 0.333 0.353 0.298 Population sizea, b, c 0.231 0.132 0.205 0.097 0.403 0.013 0.591 0.001 0.289 0.033 0.153 0.123 0.163 0.177 0.135 Oil exporting countriesa 0.493 0.435 0.931 0.044 0.579 0.354 1.185 0.029 0.429 0.433 0.630 0.463 0.625 0.541 0.547 % mountaineousb 0.018 0.053 0.005 0.516 0.014 0.121 0.009 0.370 0.007 0.467 0.009 0.008 0.009 0.010 0.009 Ethnic fractionalizationb -0.446 0.590 -0.402 0.566 -0.185 0.827 0.099 0.907 0.220 0.780 0.829 0.701 0.848 0.847 0.788 Religious fractionalizationb 0.981 0.414 0.517 0.568 1.489 0.175 0.020 0.986 1.392 0.150 1.201 0.906 1.099 1.123 0.968 Constant -4.334 0.000 -3.576 0.000 -4.796 0.000 -5.185 0.000 -4.503 0.000 0.685 0.476 0.686 0.818 0.586 N 1737 2195 1917 1309 1983 Likelihood-Ratio chi² (11) 30.52 60.99 55.33 37.56 49.09 Pseudo R² 0.1294 0.1624 0.201 0.1471 0.1537 Note: standard errors are in parentheses; coefficients with p < .1 are in bold letters a lagged by one year b centered on the mean c natural log taken
70
R E F E R E N C E S
Apodaca, Clair. 1998. “Measuring Women’s Economic and Social Rights Achievement.” Human Rights Quarterly 20(1): 139-72.
Apodaca, Clair. 2001. “Global Economic Patterns and Personal Integrity Rights After the Cold War.” International Studies Quarterly 45: 587-602.
Azam, Jean-Paul, and Anke Hoeffler. 2002. “Violence Against Civilians in Civil Wars: Looting or Terror?” Journal of Peace Research 39(4): 461-85.
Berejikian, Jeffrey. 1992. “Revolutionary Collective Action and the Agent-Structure Problem.” American Political Science Review 86(3): 647-57.
Boswell, Terry, and William J. Dixon. 1990. “Dependency and Rebellion: A Cross-National Analysis.” American Sociological Review 55(4): 540-59.
Braumoeller, Bear F. 2004. “Hypothesis Testing and Multiplicative Interaction Terms.” International Organization 58(Fall): 807-20.
Brecher, Michael, Patrick James, and Jonathan Wilkenfeld. 2000. “Escalation and War in the Twentieth Century – Findings from the International Crisis Behavior Project.” In What Do We Know About War? ed. John A. Vasquez. Lanham: Rowman and Littlefield: 37-53.
Carey, Sabine C. 2004. “Domestic Threat and Repression: An Analysis of State Responses to Different Forms of Dissent.” In Understanding Human Rights Violations: New Systematic Approaches, ed. Sabine C. Carey and Steven C. Poe. Aldershort: Ashgate: 207-25.
Carment, David. 1993. “The International Dimensions of Ethnic Conflict: Concepts, Indicators, and Theory.” Journal of Peace Research 30(2): 137-50.
Carment, David. 2003. “Assessing state failure: implications for theory and policy.” Third World Quarterly 24(3): 407-27.
Chong, Dennis. 1991. Collective Action and the Civil Rights Movement. Chicago: University of Chicago Press.
Cingranelli, David L., and David L. Richards. 1999. “Measuring the Level, Pattern, and Sequence of Government Respect for Physical Integrity Rights.” International Studies Quarterly 43: 407-17.
Cingranelli, David L., and David L. Richards. 2004. The Cingranelli-Richards (CIRI) Human Rights Database – Coder Manual. http://ciri.binghamton.edu/index.asp, accessed October 2004.
71
Collier, Paul, and Anke Hoeffler. 2001. “Greed and Grievance in Civil War.” World Bank working paper no. 2355. Washington DC: World Bank.
http://www.worldbank.org/research/conflict/papers/greedandgrievance.htm, accessed March 2005.
Collier, Paul, Anke Hoeffler, and Måns Söderbom. 2004. “On the Duration of Civil War.” Journal of Peace Research 41(3): 253-73.
Davenport, Christian. 1995. “Multi-Dimensional Threat Perception and State Repression: An Inquiry Into Why States Apply Negative Sanctions.” American Journal of Political Science 39(3): 683-713.
Davenport, Christian. 1996a. “The Weight of the Past: Exploring Lagged Determinants of Political Repression.” Political Research Quarterly 49(2): 377-403.
Davenport, Christian. (1996b. “"Constitutional Promises" and Repressive Reality: A Cross-National Time-Series Investigation of Why Political and Civil Liberties are Suppressed.” Journal of Politics 58(3): 627-54.
Davenport, Christian, Will H. Moore, and Steven C. Poe. 2003. “Sometimes You Just Have To Leave: Domestic Threats and Forced Migration, 1964-1989.” International Interactions 29: 27-55.
Davenport, Christian. 2004. “The Promise of Democratic Pacification: An Empirical Assessment.” International Studies Quarterly 48(3): 539-60.
Davenport, Christian, and David A. Armstrong II. 2004. “Democracy and the Violation of Human Rights: A Statistical Analysis from 1976 to 1996.” American Journal of Political Science 48(July): 538-54.
Davies, James C. 1962. “Toward a Theory of Revolution.” American Sociological Review 27(1): 5-19.
Davies, John L., and Ted Robert Gurr, eds. 1998. Preventive Measures – Building Risk Assessment and Crisis Early Warning Systems. Lanham: Rowman; & Littlefield.
Doyle, Michael W., and Nicholas Sambanis. 2000. “International Peacekeeping: A Theoretical and Quantitative Analysis.” American Political Science Review 94(4): 779-801.
Elbadawi, Ibrahim, and Nicholas Sambanis. 2002. “How Much War Will We See? Explaining the Prevalence of Civil War.” Journal of Conflict Resolution 46(3): 307-34.
72
Fearon, James D., and David D. Laitin. 1996. “Explaining Interethnic Cooperation.” American Political Science Review 90(4): 715-35.
Fearon, James D., and David D. Laitin. 2003a. “Ethnicity, Insurgency, and Civil War.” American Political Science Review 97(3): 75-90.
Fearon, James D., and David D. Laitin. 2003b. “Additional Tables for “Ethnicity, Insurgency, and Civil War.”” http://www.stanford.edu/group/ethnic/, accessed November 2004.
Fearon, James D. 2004. “Why Do Some Civil Wars Last So Much Longer Than Others?” Journal of Peace Research 41(3): 275-301.
Fein, Helen. 1995. “More Murder in the Middle: Life-Integrity Violations and Democracy in the World, 1987.” Human Rights Quarterly 17(1): 170-91.
Gartner, Scott Sigmund, and Patrick M. Regan. 1996. “Threat and Repression: The Non-Linear Relationship between Government and Opposition Violence.” Journal of Peace Research 33(3): 273-87.
Gates, Scott. 2002. “Recruitment and Allegiance – The Microfoundations of Rebellion.” Journal of Conflict Resolution 46(1): 111-30.
Ghobarah, Hazem Adam, Paul Huth, and Bruce Russett. 2003. “Civil Wars Kill and Maim People - Long After the Shooting Stops.” American Political Science Review 97(2): 189-202.
Gibney, Mark. 2004. Political Terror Scale. http://www.unca.edu/politicalscience/ faculty-staff/gibney.html, accessed October 2004.
Gleditsch, Nils Petter, Håvard Strand, Mikael Eriksson, Margareta Sollenberg, and Peter Wallensteen. 2001. “Armed conflict 1945-99: A new dataset.” Unpublished paper, PRIO, Oslo Norway [adopted from Sambanis 2004].
Goldstone, Jack A., Ted Robert Gurr, Barbara Harff, Marc A. Levy, Monty G. Marshall, Robert H. Bates, David L. Epstein, Colin H. Kahl, Pamela T. Surko, John C. Ulfelder, and Alan N. Unger. 2000. “State Failure Task Force Project: Phase III Findings.” http://www.cidcm.umd.edu/inscr/stfail/SFTF%20Phase% 20III%20Report%20Final.pdf, accessed August 2003.
Gupta, Dipak K., Harinder Singh, and Tom Sprague. 1993. “Government Coercion of Dissidents: Deterrence or Provocation?” Journal of Conflict Resolution 37(2): 301-39.
Gurr, Ted Robert. 1968. “Psychological Factors in Civil Violence.” World Politics 20: 245-78.
73
Gurr, Ted Robert. 1970. Why Men Rebel. Princeton: Princeton University Press.
Gurr, Ted Robert, and Will H. Moore. 1997. “Ethnopolitical Rebellion: A Cross-Sectional Analysis of the 1980s with Risk Assessments for the 1990s.” American Journal of Political Science 41: 1079-103.
Harff, Barbara, and Ted Robert Gurr. 1988. “Toward Empirical Theory of Genocides and Politicides: Identification and Measurement of Cases Since 1945.” International Studies Quarterly 32(3): 359-71.
Harff, Barbara. 2003. “No Lessons Learned from the Holocaust? Assessing Risks of Genocide and Political Mass Murder since 1955.” American Political Science Review 97(1): 57-73.
Hathaway, Oona A. 2002. “Do Human Rights Treaties Make a Difference?” Yale Law Journal 111(8): 1935-2042.
Hegre, Håvard, Tanja Ellingsen, Scott Gates, and Nils Petter Gleditsch. 2001. “Toward a Democratic Civil Peace? Democracy, Political Change, and Civil War, 1816-1992.” American Political Science Review 95(1): 33-48.
Henderson, Conway W. 1991. “Conditions Affecting the Use of Political Repression.” Journal of Conflict Resolution 35(1): 120-42.
Henderson, Conway W. 1993. “Population Pressures and Political Repression.” Social Science Quarterly 74: 322-33.
Horowitz, Donald L. 1985. Ethnic Groups in Conflict. Berkeley: University of California Press.
Kaufman, Stuart J. 1996. “Spiraling to Ethnic War – Elites, Masses, and Moscow in Moldova’s Civil War.” International Security 21(2): 108-38.
Kaufmann, Chaim. 1996. “Possible and Impossible Solutions to Ethnic Civil Wars.” International Security 20(4): 136-75.
Keith, Linda Camp. 1999. “The United Nations International Covenant on Civil and Political Rights: Does It Make A Difference in Human Rights Behavior?” Journal of Peace Research 36(1): 95-118.
Keith, Linda Camp. 2002. “Constitutional Provisions for Individual Human Rights (1977-1996): Are They More than Mere “Window Dressing?”” Political Research Quarterly 55(1): 111-43.
74
Keith, Linda Camp, and Steven C. Poe. 2004. “Are Constitutional State of Emergency Clauses Effective? An Empirical Exploration.” Human Rights Quarterly 26: 1071-97.
Krain, Matthew. 1997. “State-Sponsored Mass Murder: The Onset and Severity of Genocides and Politicides.” Journal of Conflict Resolution 41(3): 331-60.
Krain, Matthew, and Marissa E. Myers. 1997. “Democracy and Civil War: A Note on the Democratic Peace Proposition.” International Interactions 23: 87-94.
Lacina, Bethany. 2005. “Understanding and Explaining the Severity of Civil Wars.” Paper presented at the Annual Convention of the International Studies Association, March 1 – 5, 2005, in Honolulu, Hawaii.
Lake, David A., and Donald Rothchild. 1996. “Containing Fear – The Origins and Management of Ethnic Conflict.” International Security 21(2): 41-75.
Lichbach, Mark Irving. 1987. “Deterrence or Escalation? The Puzzle of Aggregate Studies of Repression and Dissent.” Journal of Conflict Resolution 31(2): 266-97.
Lichbach, Mark Irving. 1995. The Rebel’s Dilemma. Ann Arbor: University of Michigan Press.
Madani, Hamed. 1992. “Socioeconomic Development and Military Policy Consequences of Third World Military and Civilian Regimes, 1965-1985.” Ph.D. diss. University of North Texas.
Mansfield, Edward D., and Jack Snyder. 2002. “Democratic Transitions, Institutional Strength, and War.” International Organization 56(2): 297-337.
Marshall, Monty G., and Keith Jaggers. 2002. Polity IV Project – Political Regime Characteristics and Transitions, 1800-2002 – Dataset User’s Manual. www.cidcm.umd.edu/inscr/polity, accessed October 2004.
Mason, T. David, and D. Krane. 1989. “The Political Economy of Death Squads: Toward a Theory of the Impact of State-Sanctioned Terror.” International Studies Quarterly 33: 175-98.
McCormick, James M., and Neil J. Mitchell. 1997. “Human Rights Violations, Umbrella Concepts, and Empirical Analysis.” World Politics 49(4): 510-25.
Midlarsky, Manus I. 1988. “Rulers and the Ruled: Patterned Inequality and the Onset of Mass Political Violence.” American Political Science Review 82(2): 491-509.
Mitchell, Neil J., and James M. McCormick. 1988. “Economic and Political Explanations of Human Rights Violations.” World Politics 40(4): 476-98.
75
Moore, Will H. 1995. “Rational Rebels: Overcoming the Free-Rider Problem.” Political Research Quarterly 48(2): 417-54.
Moore, Will H. 1998. “Repression and Dissent: Substitution, Context, and Timing.” American Journal of Political Science 42(3): 851-73.
Moore, Will H., and Stephen M. Shellman. 2004. “Fear of Persecution: Forced Migration, 1952-1995.” Journal of Conflict Resolution 40(5): 723-45.
Most, Benjamin A., and Harvey Starr. 1989. Inquiry, Logic and International Politics. Columbia, SC: University of South Carolina Press
Mueller, John. 2000. “The Banality of “Ethnic War.”” International Security 25(1): 42-70.
Muller, Edward N. 1985. “Income Inequality, Regime Repressiveness, and Political Violence.” American Sociological Review 50(1): 47-61.
Muller, Edward N., and Mitchell A. Seligson. 1987. “Inequality and Insurgency.” American Political Science Review 81(2): 425-52.
Muller, Edward N., and Erich Weede. 1990. “Cross-National Variation in Political Violence: A Rational Action Approach.” Journal of Conflict Resolution 34(4): 624-51.
Oliver, Pamela. 1980. “Rewards and Punishments as Selective Incentives for Collective Action: Theoretical Investigations.” American Journal of Sociology 85(6): 1365-75.
Olson, Mancur 1965. The Logic of Collective Action. Cambridge: Harvard University Press.
Pion-Berlin, David, and George A. Lopez 1991. “Of Victims and Executioners: Argentine State Terror, 1975-1979.” International Studies Quarterly 35(1): 63-86.
Poe, Steven C., and C. Neal Tate. 1994. “Repression of Human Rights to Personal Integrity in the 1980s: A Global Analysis.” American Political Science Review 88(4): 853-72.
Poe, Steven C., Dierdre Wendel-Blunt, and Karl Ho. 1997. “Global Patterns in the Achievement of Women's Human Rights to Equality.” Human Rights Quarterly 19(4): 813-35.
Poe, Steven C., C. Neal Tate, and Linda Camp Keith. 1999. “Repression of the Human Right to Personal Integrity Revisited: A Global Cross-National Study Covering the Years 1976-1993.” International Studies Quarterly 43: 291-313.
76
Poe, Steven C., Sabine C. Carey, and Tanya C. Vazquez. 2001. “How are These Pictures Different? A Quantitative Comparison of the US State Department and Amnesty International Human Rights Reports, 1976-1995.” Human Rights Quarterly 23: 650-77.
Poe, Steven C. 2004. “The Decision to Repress: An Integrative Theoretical Approach to the Research on Human Rights and Repression.” In Understanding Human Rights Violations: New Systematic Approaches, ed. Sabine C. Carey and Steven C. Poe. Aldershort: Ashgate: 17-39.
Poe, Steven C., Nicolas Rost, and Sabine C. Carey. 2005. “Recognizing Regimes Risking Repression and Ripe for Rights Realization: Toward A Systematic Risk (and Opportunity) Assessment for Physical Integrity Rights.” Paper presented at the Annual Convention of the International Studies Association, March 1 – 5, 2005, in Honolulu, Hawaii.
Regan, Patrick M. 1995. “U.S. Economic Aid and Political Repression: An Empirical Evaluation of U.S. Foreign Policy.” Political Research Quarterly 48(3): 613-28.
Regan, Patrick M. 1996. “Conditions of Successful Third-Party Intervention in Intrastate Conflicts.” Journal of Conflict Resolution 40(2): 336-59.
Regan, Patrick M., and Errol A. Henderson. 2002. “Democracy, threats and political repression in developing countries: are democracies internally less violent?” Third World Quarterly 23(1): 119-36.
Richards, David L., Ronald Gelleny, and David H. Sacko. 2001. “Money with a Mean Streak? Foreign Economic Penetration and Government Respect for Human Rights in Developing Countries.” International Studies Quarterly 45: 219-39.
Sambanis, Nicholas. 2004. “What Is Civil War? Conceptual and Empirical Complexities of an Operational Definition.” Journal of Conflict Resolution 48(6): 814-58.
Sarkees, Meredith Reid, and J. David Singer. 2001. “The Correlates of War datasets: The totality of war.” Paper prepared for the 42nd Annual Convention of the International Studies Association, February 20-24, Chicago [adopted from Sambanis 2004].
Schmeidl, Susanne. 1997. “Exploring the Causes of Forced Migration: A Pooled Times-Series Analysis, 1971-1990.” Social Science Quarterly 78(2): 284-308.
Schock, Kurt. 1996. “A Conjunctural Model of Political Conflict: The Impact of Political Opportunities on the Relationship between Economic Inequality and Violent Political Conflict.” Journal of Conflict Resolution 40(1): 98-113.
77
Skocpol, Theda. 1979. States and Social Revolutions – A Comparative Analysis of France, Russia, and China. Cambridge, Massachusetts: Cambridge University Press.
Snyder, Jack. 2000. From Voting to Violence – Democratization and Nationalist Conflict. New York: Norton.
Taylor, Michael. 1988. “Rationality and revolutionary collective action.” In Rationality and Revolution, ed. Michael Taylor. Cambridge, UK: Cambridge University Press.
Tilly, Charles. 1978. From Mobilization to Revolution. New York: McGraw-Hill.
United Nations. 2004. A more secure world: our shared responsibility – Report of the High-level Panel on Threats, Challenges and Change.
www.un.org/ secureworld, accessed December 2004.
Valentino, Benjamin, Paul Huth, and Dylan Balch-Lindsay. 2004. ““Draining the Sea”: Mass Killing and Guerilla Warfare.” International Organization 58: 375-407.
Van Belle, Douglas. 1996. “Leadership and Collective Action: The Case of Revolution.” International Studies Quarterly 40: 107-32.
Walker, Scott, and Steven C. Poe. 2002. “Does Cultural Diversity Affect Countries' Respect for Human Rights?” Human Rights Quarterly 24: 237-63.
Wantchekon, Leonard. 2004. “The Paradox of "Warlord" Democracy: A Theoretical Investigation.” American Political Science Review 98(1): 17-33.
Zanger, Sabine. 2000. “A Global Analysis of the Effect of Political Regime Changes on Life Integrity Violations, 1977-1993.” Journal of Peace Research 37(2): 213-33.
Zartman, I. William. 1995. “Dynamics and Constraints in Negotiations in Internal Conflicts.” Introduction in Elusive Peace – Negotiating an End to Civil Wars, ed. I. William Zartman. Washington DC: Brookings Institution: 3-29.