Journal of Peace Research

http://jpr.sagepub.com/

Post-Conflict Risks
Paul Collier, Anke Hoeffler and Måns Söderbom
Journal of Peace Research 2008 45: 461
DOI: 10.1177/0022343308091356

The online version of this article can be found at:

http://jpr.sagepub.com/content/45/4/461

Published by:

http://www.sagepublications.com

On behalf of:
Peace Research Institute Oslo

Journal of Peace Research Replication Data

Additional services and information for Journal of Peace Research can be found at:

Email Alerts: http://jpr.sagepub.com/cgi/alerts

Subscriptions: http://jpr.sagepub.com/subscriptions

Reprints: http://www.sagepub.com/journalsReprints.nav

Permissions: http://www.sagepub.com/journalsPermissions.nav

Citations: http://jpr.sagepub.com/content/45/4/461.refs.html

>> Version of Record - Jun 24, 2008

What is This?

Downloaded from jpr.sagepub.com at UNIV TORONTO on November 15, 2014

 © 2008 Journal of Peace Research,
vol. 45, no. 4, 2008, pp. 461–478
Sage Publications (Los Angeles, London, New Delhi
and Singapore) http://jpr.sagepub.com
DOI 10.1177/0022343308091356

Post-Conflict Risks*
PAUL COLLIER, ANKE HOEFFLER
Centre for the Study of African Economies, Department of Economics,
University of Oxford

MÅNS SÖDERBOM
Department of Economics, School of Business, Economics and Law, University
of Gothenburg

Post-conflict societies face two distinctive challenges: economic recovery and reduction of the risk of a
recurring conflict. Aid and policy reforms have been found to be effective in economic recovery. In this
article, the authors concentrate on the other challenge – risk reduction. The post-conflict peace is typ-
ically fragile: nearly half of all civil wars are due to post-conflict relapses. The authors find that eco-
nomic development substantially reduces risks, but it takes a long time. They also find evidence that
UN peacekeeping expenditures significantly reduce the risk of renewed war. The effect is large: dou-
bling expenditure reduces the risk from 40% to 31%. In contrast to these results, the authors cannot
find any systematic influence of elections on the reduction of war risk. Therefore, post-conflict elec-
tions should be promoted as intrinsically desirable rather than as mechanisms for increasing the dura-
bility of the post-conflict peace. Based on these results, the authors suggest that peace appears to depend
upon an external military presence sustaining a gradual economic recovery, with political design playing
a somewhat subsidiary role. Since there is a relationship between the severity of post-conflict risks and
the level of income at the end of the conflict, this provides a clear and uncontroversial principle for
resource allocation: resources per capita should be approximately inversely proportional to the level of
income in the post-conflict country.

Introduction Supporting the recovery was the founding

objective of the World Bank, and both aid
Post-conflict societies face two distinctive
and policy reforms have been found to be
challenges: economic recovery and reduction
highly effective in the post-conflict context
of the risk of a recurring conflict. Conflict will
(Collier & Hoeffler, 2004a). In this article,
usually have severely damaged the economy.
we study the other challenge, reduction of the

* We thank the Guest Editors of the current issue of Journal

of Peace Research, Ibrahim Elbadawi, Håvard Hegre and at the OECD for comments on previous drafts. The
Gary Milante, and three anonymous referees for very research is supported by the New Security Challenges
helpful comments on an earlier version of the article. We Programme of the Economic and Social Research Council
would like to thank Sarah Cliffe for suggesting the analy- and by the Political Institutions, Development, and
sis of UN military peacekeeping expenditures and for Domestic Civil Peace Programme of the Government of
arranging access to the data, and Meron Destra and Alice Norway and the World Bank. All errors are our own.
Kuegler for transforming them into a manageable form. We Correspondence: anke.hoeffler@economics.ox.ac.uk. The
would like to thank participants at seminars at the data are available from http://www.prio.no/datasets and
International Peace Research Institute, Oslo (PRIO), and http://users.ox.ac.uk/~ball0144.

Downloaded from jpr.sagepub.com at UNIV TORONTO on November 15, 2014 461

462 jour nal of P E A C E R E S E A RC H volume 45 / number 4 / july 2008

risk of a recurring conflict. The post-conflict Post-Conflict Risks: Current Policy

peace is typically fragile: around half of all and Recent Research
civil wars are due to post-conflict relapses
Post-conflict policy has arisen as a practical
(Collier et al., 2003). Both external actors and
international concern only since the end of
the post-conflict government must therefore
the Cold War. Since then, very substantial
give priority to reducing the risk of conflict.
international resources have been devoted to
The two objectives of economic recovery and
post-conflict situations, often under intense
risk reduction are likely to be complemen-
media scrutiny. The predominant learning
tary: economic recovery may reduce risks,
process has been practitioner-based. Because
and risk reduction may speed recovery.
it is a recent phenomenon, academic research
However, this complementarity between
has taken time to address the subject, and
objectives does not imply coincidence of
inevitably until recently, the only feasible
instruments: the instruments that are effec-
approach was through case studies. Paris
tive for risk reduction may be quite distinct
(2004) and Stedman, Rothchild & Cousens
from those for economic recovery.
(2003) provide a comprehensive overview of
Although there is now a large case study
peacebuilding during the 1990s. Only
literature on post-conflict situations, there are
recently has an academic literature developed
few quantitative comparative studies of post-
based on theory and quantitative analysis.
conflict risks. Indeed, until recently, there
We first review practitioner learning and
were insufficient data to support such a study.
then turn to the academic literature.
Our approach is to estimate hazard functions
on a comprehensive sample of post-conflict
situations. In the next section, we discuss the Current Policy Models
hypotheses that underpin current interna- Until the end of the Cold War, the interna-
tional practice in post-conflict situations. tional community was not in a position to
These hypotheses are not explicitly derived intervene in post-conflict situations. As a
from political or economic theory, but rather consequence, there was little call for analysis.
have emerged over the past 15 years of prac- With the end of the Cold War, numerous
titioner experience. We suggest that, to an opportunities for intervention rapidly arose,
extent, they contrast with theory-based and this yielded a phase in which practition-
hypotheses. In the subsequent section, we ers had little choice but to learn by doing.
discuss our methodology and our sources of One of the conventional wisdoms that
data. There are various, often difficult choices emerged from this experience was that there
that must be made in building a dataset of were enormous differences among post-con-
post-conflict experiences that is comprehen- flict situations, so that any generalization was
sive, up to date and satisfactorily delineated dangerous. However, the decision of the
so as to exclude circumstances that do not United Nations in September 2005 to estab-
really reflect the end of a civil war. In the lish a Peacebuilding Commission reflected a
penultimate section, we present our results, growing sense that it is time to introduce a
showing how the hazard of conflict reversion greater degree of standardization. For ex-
is affected by both the initial post-conflict ample, one by-product of the ‘every situation
characteristics and time-varying characteris- is unique’ doctrine was that the amount
tics. The final section concludes with a dis- spent on peacekeeping forces has varied
cussion of the implications for international enormously between different post-conflict
policy. situations with little apparent relation to

Downloaded from jpr.sagepub.com at UNIV TORONTO on November 15, 2014

 Paul Collier et al. P O S T-C O N F L I C T R I S K S 463

need. In the absence of fresh evidence, the levels of governance and economic policy.
most likely outcome is that the Peace- For example, on the World Bank’s five-point
building Commission will itself attempt to rating system, the Country Policy and
codify current practices. Thus, at the risk of Institutional Assessment, the typical post-
imposing a greater commonality on practice conflict country starts with a rating of only
than exists, it is useful to suggest what such 2.41, a level far below the minimum level
a codification is likely to yield in terms of a regarded as adequate for development
few rules of thumb. At a minimum, an effort (Collier & Hoeffler, 2004a). Since post-
to summarize conventional wisdom (see, for conflict situations are typically characterized
example, Ashdown, 2007) into a few rules of by power struggles, the reform of economic
thumb may, through inaccuracies, provoke a policy is seldom treated as a high priority by
more authoritative succinct statement from governments themselves.
practitioners themselves. The fifth stage is to gain acceptance for
The first stage in the sequence of interna- the settlement through post-conflict elections.
tional post-conflict practice starts while the Elections are seen as legitimating both the
conflict is still continuing. This is to achieve settlement itself and the authority of the gov-
a negotiated settlement rather than allow ernment and so help to reduce tensions.
the conflict to drag on to the point of victory The sixth stage is the withdrawal of inter-
for one or other of the parties. This has been national peacekeeping troops. Sometimes the
a highly successful strategy, yielding resolu- election is treated as the milestone for this
tions to several wars. withdrawal. Just as international inter-
The second stage is a relatively light vention is seen as pump-priming democracy,
presence of peacekeeping troops, the typical the election is seen as pump-priming peace,
level of such expenses where such troops are with initial tensions being swiftly reduced by
provided being around $5 per head of the time.
population. These conventional rules of thumb guide
The third stage is to encourage a constitu- our empirical enquiry. We will attempt to
tion which provides for a degree of democracy. test whether international peacekeeping
Intervention is seen essentially as pump- troops are effective, and we will try to get
priming democracy. The degree of decentral- some measure of appropriate levels of provi-
ization of power envisaged in constitution sion. We will try to assess the extent to which
varies. The settlement in Sudan provides con- pump-priming democracy is likely to
siderable autonomy for the South, but more strengthen post-conflict peace, and the con-
usually it favours a fairly unified state. sequences of different degrees of decentral-
The fourth stage is that during its period ization. We will measure whether economic
of presence in the country, international issues can safely be left on the back-burner
intervention should be conducted in such a while more pressing political issues are
way as to leave a ‘light footprint’. A key pursued. Finally, we will attempt to assess
implication is to allow the pace of reform to whether elections are effective in strengthen-
be set by the government. This is partly ing peace.
due to sensitivity over issues of sovereignty Going beyond these rules of thumb,
and partly due to a concern that, since post- we will also try to get some sense of how
conflict situations are fragile, reform could post-conflict risks vary according to simple
easily be destabilizing. Most post-conflict and observable initial characteristics that
situations commence with extremely poor the Peacebuilding Commission could use,

Downloaded from jpr.sagepub.com at UNIV TORONTO on November 15, 2014

464 jour nal of P E A C E R E S E A RC H volume 45 / number 4 / july 2008

without controversy, to allocate the resources resources can provide rebel organizations
available for post-conflict situations more with finance. While the significance of most
effectively. of these variables could also be explained in
other terms, feasibility provides a simple and
Theory and Quantification unifying account.
There is as yet no academic consensus on the There has been little application of this
causes of violent conflict, let alone an agreed work to the post-conflict situation. Within
theory that is specific to post-conflict. How- the framework of their logit model of
ever, the predominant recent theoretical the causes of conflict, Collier & Hoeffler
position, as described in the survey of civil (2004b) investigated the effect of the passage
war in the Handbook of Defense Economics, of time since the most recent previous con-
has been to emphasize that the conditions flict. They found that risks fall with time,
that determine the feasibility of rebellion implying that the post-conflict decade is
are more important than those that influ- unusually risky. This high risk is indeed con-
ence motivation (Collier & Hoeffler, 2007a). sistent with the feasibility thesis. By revela-
The defining feature of civil war is the emer- tion, post-conflict societies are societies in
gence and durability of a private rebel army, which rebellion had proved to be feasible.
and, under most conditions, such organiza- This is somewhat analogous to the celebrated
tions are likely to be neither financially nor prediction from the economic theory of
militarily feasible. Somewhat analogous to crime that criminals would have a high rate
Hirshleifer’s ‘Machiavelli Theorem’ (Hirshleifer, of recidivism (Becker, 1968). Further, during
2001), the feasibility thesis suggests that conflict, one, or often several, rebel organi-
where insurrection is feasible, it will occur, zations are assembled, and stocks of arma-
with the actual agenda of the rebel movement ments are amassed. Both of these are legacies
being indeterminate. There is now reasonable that lower the cost of rebellion in the post-
empirical support for this thesis. Recent conflict period.
quantitative studies of the causes of civil war The feasibility theory can be contrasted
find that variables that are most readily inter- with the current-policy model. As discussed
preted as indicators of feasibility are import- above, currently policy addresses post-conflict
ant, namely, mountainous terrain, protection risks primarily through political design.
through external security commitments, low Underlying this is an implicit theory of the
per capita income, slow economic growth causes of conflict which gives precedence to
and large exports of natural resources motivation and, in particular, to grievances
(Fearon & Laitin, 2003; Miguel, Satyanath & based on political exclusion. Walter (1999)
Sergenti, 2004; Collier & Hoeffler, 2004b; and Hartzell & Hoddie (2003) challenge this
Collier, Hoeffler & Rohner, 2007). Moun- view by examining a subset of post-conflict
tainous terrain provides safe havens for rebel situations, namely, those conflicts that ended
forces, whereas external security commit- in a negotiated settlement. Their analysis
ments may discourage rebellion by reducing suggests that credible guarantees on the
the prospects of success. Low per capita terms of the agreement, especially in form of
income is interpreted by Fearon & Laitin as international security arrangements, make it
proxying the incapacity of the state to main- more likely for the peace to last.
tain effective control over its territory. Both If, indeed, feasibility and credibility are the
low income and slow growth can be inter- decisive factors, then economic and military
preted as lowering the recruitment cost of instruments might be more important in
rebel troops, and the predation of natural post-conflict situations. To date, the only

Downloaded from jpr.sagepub.com at UNIV TORONTO on November 15, 2014

 Paul Collier et al. P O S T-C O N F L I C T R I S K S 465

aspect of general post-conflict risk that has explanatory variable xtj leads to an increase in
been modelled is the effect of military spend- the hazard of war and a reduction in the
ing by the government, which has been found expected duration of peace; and vice versa if
to have a differential effect in post-conflict sit- ␤j ⬍0. For the baseline hazard h B(t), we adopt
uations (Collier & Hoeffler, 2006, 2007b). a piecewise exponential model, which is quite
Typically, post-conflict governments main- flexible. Our starting point is a specification
tain military spending at a very high level, in where we divide the time axis into W intervals
part as a response to the high risk of further by the points c1,c2, …, and assume constant
conflict. However, uniquely in post-conflict baseline hazard rates within each interval:
situations, allowing for the interdependence
⎛ W
⎞
between risks and spending, such spending is h B (t )⫽ exp ⎜
⫹ ∑ ␭w d w ( t ) ⎟ , (2)
counterproductive. Collier & Hoeffler attempt ⎝ W ⫽2 ⎠
to explain this through a game-theoretic where dw (t) is a duration dummy variable
model in which high military spending by the equal to 1 if Cw ⫺1 ⬍t ⱕCw for C0 ⫽0 and
government inadvertently signals an inten- Cw ⫽⬁, and 0 otherwise; ␣ is an intercept;
tion on the part of the government to exploit and ␭2,…,␭w are baseline hazard parameters.
the potential time-inconsistency of any peace Thus, the baseline hazard is allowed to vary
agreement, by reneging on the terms of freely between intervals, which imposes few
settlement. restrictions on duration dependence. We
include an intercept in the model and
exclude the first duration dummy (i.e. d1),
Methodology and Data
which implies that negative (positive) coeffi-
The approach adopted in the previous empir- cients on ␭2,…,␭w imply that the hazard is
ical analyses of the causes of conflict discussed lower (higher) than in the first interval.
above was that of logit analysis or its variants. A useful metric in discussing post-conflict
Such an approach cannot investigate in any risks is the average risk that the peace will col-
depth either the distinctive structure of post- lapse. In our sample of 68 post-conflict
conflict risks or how they evolve as a result of episodes, 31 reverted to war, so that the
policy choices. In this article, our approach is average risk was 46%. Since our focus is
to estimate a hazard function of the risk of going to be on the first post-conflict decade,
conflict reversion on a sample confined to it is useful to consider those reversions that
post-conflict countries. We assume the hazard occurred during that decade. The standard
is exponential and proportional: non-parametric Kaplan-Meier estimator
indicates that the likelihood of survival for at
h(x ␶ , ␤;t) ⫽exp(x␶ , ␤)hB(t), (1) least 10 years is 0.60. Hence, the implied
likelihood of failure (i.e. reversion to war) in
where t denotes the duration of a post-conflict the first decade of post-conflict peace is 40%.
peace period, x␶ is a vector of exogenous vari- This serves as a benchmark for calibrating
ables observed at calendar time ␶ , ␤ is a vector the importance of each explanatory variable.
of unknown parameters and hB is the baseline With our modelling framework, the likeli-
hazard.1 With this specification, ␤j ⬎ 0 hood that a country ‘survives’ the first decade
implies that an increase in the associated of peace is given by the survival function
evaluated at t ⫽10 years:
⎛ 10 ⎞
We use a ␶ subscript on the vector of explanatory vari- S (10 ) = exp ⎜⫺ ∫ h ( .; u ) du ⎟ ,
1

ables to indicate that the explanatory variables may be, but (3)
are not necessarily, time-varying.
⎝ u⫽0 ⎠

Downloaded from jpr.sagepub.com at UNIV TORONTO on November 15, 2014

466 jour nal of P E A C E R E S E A RC H volume 45 / number 4 / july 2008

and so the risk of a collapse is given by threshold is used to define the restoration of
F(10) ⫽1 ⫺S(10). peace, then the episode might be treated as a
The limitation of the hazard approach is continuous war. Thus, the problems with
that, by dint of the sample, it cannot make respect to dating the start and end of the con-
comparisons either with risks in countries flict are of importance not only for the analy-
that are not post-conflict or with risks that sis of the duration of conflicts but also for the
prevailed in the post-conflict country prior to analysis of post-conflict risks.
conflict. However, it enables us to be much The two most commonly used datasets are
more precise in our investigation of what the Correlates of War (COW) project, initi-
determines the initial risk of conflict rever- ated by Small & Singer (1982) and updated
sion, how that risk evolves naturally simply by by Gleditsch (2004), and the Uppsala/PRIO
the passage of time, and how its evolution is Armed Conflict Dataset (ACD) by Gleditsch
affected by interventions during the post- et al. (2002). The COW definition of civil
conflict period. We have previously used this wars is based on four main characteristics. It
approach to investigate the duration of civil requires that there is organized military action
war (Collier, Hoeffler & Söderbom, 2004). and that at least 1,000 battle-deaths resulted.
For the analysis of post-conflict risks, we In order to distinguish wars from genocides,
need to date both the end of conflict and, massacres and pogroms, there has to be effec-
should the society revert to conflict, the end tive resistance – at least 5% of the deaths
of peace. Dating the start and, more impor- have been inflicted by the weaker party. The
tantly for this article, the end of the conflict definition of war as used in the ACD has two
is often difficult. Trigger events can be dated, main dimensions. First, they distinguish four
but often the violence escalates over some types of violent conflicts according to the par-
period of time before it reaches the relevant ticipants and location: (1) extra-systemic con-
threshold for it to be classified as a civil war. flicts (essentially colonial or imperialist wars),
Wars end with a military victory, settlement (2) interstate wars, (3) intrastate wars and (4)
or truce. About half of all civil wars end in internationalized intrastate wars. The second
the military defeat of one party (Sambanis, dimension defines the level of violence: minor
2000). This makes dating the ending of wars conflicts produce more than 25 battle-
somewhat easier than using the dates of related deaths per year; intermediate con-
peace agreements, some of which may not flicts produce more than 25 battle-related
have resulted in an end to military action. deaths per year and a total conflict history of
The duration of the war depends not only on more than 1,000 battle-related deaths; and
being able to date a start and end but also on wars are conflicts which result in more than
the definition of violence thresholds. Data- 1,000 battle-related deaths per year. The
sets that define a civil war by 1,000 battle- ACD does not report end dates to the con-
related deaths per year have, on average, flicts and, for the present analysis, we use the
shorter wars than datasets with lower thresh- updated COW data.2
olds. Consider, for example, a war with more Our analysis spans the 1960–2002 period.
than 1,000 battle-related deaths during the We start in 1960 simply because we wish to
first year. If the number of such deaths falls link the duration data on peace periods with
beneath the threshold in the second year but economic and political variables, most of
reaches it again during the third year, a rigid
application of the 1,000-deaths criterion
leads to the episode being classified as a failed 2 Data are described in Gleditsch (2004) and are available
post-conflict situation. If, however, a lower from http://weber.ucsd.edu/~kgledits/expwar.html.

Downloaded from jpr.sagepub.com at UNIV TORONTO on November 15, 2014

 Paul Collier et al. P O S T-C O N F L I C T R I S K S 467

which are not available before 1960. A risk of war falls as the time since the settle-
complete list of the peace periods used in ment increases, the coefficients on the dura-
our analysis is provided in Appendix A. As tion dummies would be negative (since the
can be seen, peace duration is defined first duration dummy is excluded). Indeed, if
simply as the length of time between the the relationship is monotonic, this would
end of a conflict and the beginning of a show up as larger and larger negatives in the
new one (this would constitute an uncen- estimates of ␭2,…,␭w. In choosing the cutoff
sored peace spell), or between the end of a points for the duration dummies and the
conflict and the end of the time period over number of duration dummies to allow for, we
which the country is observed, which is let the data guide us. We started out using a
usually 31 December 2002 (this would be a fairly rich set of duration dummies, allowing
censored spell). the hazard to change every two years during
the first decade of post-conflict peace.
Omitting the first duration dummy, d1, we
The Hazard Generated by Conflict
then tested for the joint significance of
Legacy
␭2,…,␭5 and found we could not reject
We now estimate a standard hazard function the null hypothesis that these coefficients are
on our data. Our core regression is shown in jointly zero. This suggests that the baseline
Table I, column 1. In subsequent columns, hazard may be flat, that is, that the risk of
we investigate variations. The table reports conflict does not depend on the length of
estimated coefficients (log hazard ratios) and peace, conditional on the other explanatory
the associated z-values (in parentheses). The variables in the model. We then experimented
factors that influence, or might influence, with various alternative ways of specifying the
post-conflict risks can be grouped into tem- baseline hazard, and the main result – a flat
poral, economic, political, social and mili- baseline hazard – remained robust. In the
tary. We consider the variables in these end, we decided on a very parsimonious spec-
groupings.3 ification, allowing for just one shift in the
baseline hazard, after four years of peace. All
Temporal Effects the results shown in Table I are based on this
Part of the rationale for short periods of specification. As can be seen in the table, the
peacekeeping, often merely two years, is that coefficient on the duration dummy for spells
it is the initial post-conflict period during longer than four years is negative, suggesting
which risks are highest. In our previous work a reduction in the risk of war after four years,
using logistical techniques, we found that but the effect is not large, and it is not statis-
post-conflict risks indeed fall over time. As tically significant. The risk during the first
discussed above, we include duration four years is 23%, and during the remaining
dummies in the general specification of the six years it is 17%, conditional upon the first
model in order to allow the hazard of war to four years having been peaceful. For policy
change over time. If, as one might expect, the purposes, the key conclusion is that there is
no ‘safe period’ during the decade. The entire
post-conflict decade faces a high level of risk.
3 Notice that our set of explanatory variables consists of

both time-invariant and time-varying variables. In the core

specification (column 1, Table I), all variables except ethnic Economic Influences
fractionalization and autonomy are time-varying. The The first significant economic influence is
highest frequency with which any of the explanatory vari-
ables varies is annual, e.g. log of per capita GDP and its
the level of per capita income (expressed in
growth rate. logarithmic form). To reduce the problem of

Downloaded from jpr.sagepub.com at UNIV TORONTO on November 15, 2014

468 jour nal of P E A C E R E S E A RC H volume 45 / number 4 / july 2008

Table I. Duration of Postwar Peace

(1) (2) (3) (4)

Economic
Per capita income ⫺0.427 ⫺0.431 ⫺0.423 ⫺0.551
(1.72)⫹ (1.72) ⫹ (1.70)⫹ (2.01)*
Per capita income growth ⫺3.548 ⫺3.716 ⫺3.613 ⫺4.184
(2.21)* (2.29)* (2.24)* (2.37)*
Political
Democracy 1.230 1.194 1.224 1.515
(2.43)* (2.34)* (2.42)* (2.86)**
Democracy missing (dummy) 1.752 1.726 1.754 1.798
(2.68)** (2.64)** (2.69)** (2.74)**
Regional autonomy ⫺1.561 ⫺1.619 ⫺1.538 ⫺1.148
(1.43) (1.46) (1.41) (0.99)
Regional autonomy missing (dummy) ⫺0.253 ⫺0.253 ⫺0.240 0.123
(0.50) (0.49) (0.47) (0.24)
Election shift ⫺0.709 ⫺0.754
(1.97)* (2.05)*
1st election ⫺0.495
(0.65)
Year following 1st election 0.997
(1.70)⫹
Subsequent elections ⫺0.318
(0.42)
Year following sub. Elections 0.787
(1.34)
1st election shift ⫺0.820
(1.72)⫹
Subsequent elections shift ⫺0.593
(1.21)
ln Economic freedom ⫺0.336
(1.19)
Economic freedom missing (dummy) ⫺2.757
(1.99)*
Social
ln Diaspora ⫺0.333 ⫺0.345 ⫺0.337 ⫺0.259
(2.82)** (2.86)** (2.83)** (2.10)*
Diaspora missing (dummy) 3.464 3.585 3.503 2.626
(2.46)* (2.51)* (2.48)* (1.83)⫹
Ethnic diversity ⫺1.038 ⫺1.068 ⫺1.035 ⫺1.439
(1.24) (1.27) (1.24) (1.70)⫹
Ethnic diversity missing (dummy) ⫺15.198 ⫺14.263 ⫺14.209 ⫺16.206
(0.01) (0.01) (0.01) (0.01)
Peacekeeping
ln UN peacekeeping expenditure ⫺0.405 ⫺0.414 ⫺0.407 ⫺0.478
(2.38)* (2.42)* (2.40)* (2.62)**
No UN PKO ⫺3.714 ⫺3.842 ⫺3.738 ⫺4.735
(2.16)* (2.21)* (2.18)* (2.50)*
UN data missing (dummy) ⫺3.886 ⫺3.992 ⫺3.915 ⫺4.919
(2.09)* (2.13)* (2.11)* (2.49)*

(continued)

Downloaded from jpr.sagepub.com at UNIV TORONTO on November 15, 2014

 Paul Collier et al. P O S T-C O N F L I C T R I S K S 469

Table I. (Continued)

(1) (2) (3) (4)

Time
Years 4⫹of peace ⫺0.475 ⫺0.464 ⫺0.454 ⫺0.392
(1.12) (1.03) (1.06) (0.89)
Log likelihood ⫺66.821 ⫺66.539 ⫺66.759 ⫺63.041
Number of episodes 74 74 74 74
Number of failures 33 33 33 33
Absolute values of z-statistics in parentheses. * significant at 5% level; ** significant at 1% level.

endogeneity, we introduce this variable with 2004b; Miguel, Satyanath & Sergenti, 2004;
a two-year lag. The variable is time-varying: Fearon & Laitin, 2003). However, it should
the society can do nothing about the level of be stressed that the present results are specific
income it inherits at the start of the peace, to post-conflict and generated by an entirely
but its choices will influence how the level of different statistical process from that used in
income evolves during the decade. Income the analysis of initial causes. To get a sense of
matters: it is highly significant and the effect the importance of growth, we return to the
is large. Compared with the 40% risk at the 40% benchmark risk and vary the growth
mean characteristics, if the initial level of rate. Post-conflict societies have an immensely
income is twice the mean and all other char- wide range of growth experience: outcomes
acteristics are held constant, then the decade- are radically more dispersed than in normal
risk falls to 31%. This has two powerful growth situations (Collier & Hoeffler, 2004a).
implications. One is that risks are consider- Some post-conflict economies grow rapidly,
ably higher in low-income countries: the while others continue to fall apart. If the
Democratic Republic of Congo, Liberia and economy remains stagnant through the
East Timor are all much more alarming situ- decade, the decade-risk is 42.1%. If, instead,
ations than Bosnia, other things being equal. the economy grows at 10% per year, which is
The other implication is that economic per- fast but not without precedent, the decade-
formance during the decade is likely to risk falls to 26.9%. For completeness, we
matter, if only because higher income will might note that this massive reduction is split
bring risks down. almost equally between the direct contribu-
In fact, the importance of economic per- tion of growth to risk reduction in the year in
formance is considerably greater than this, which it occurs and its cumulative contribu-
because growth itself matters.4 Faster growth tion via its legacy of higher levels of income.
directly and significantly reduces risk in the The evident implications of the economic
year in which it occurs, as well as cumulating variables is that, other things being equal,
into a higher level of income.5 These effects international post-conflict efforts should
are entirely consistent with results on the be concentrated disproportionately in the
initial causes of conflict (Collier & Hoeffler, poorest countries and should focus heavily
upon economic recovery.
A summary measure of the sort of growth-
4 Growth rates are expressed in decimals throughout the promoting policies, institutions and gover-
analysis.
5 To reduce the problem of endogeneity, we lag the growth nance favoured by the World Bank is its
rate by one year. Country Policy and Institutional Assessment

Downloaded from jpr.sagepub.com at UNIV TORONTO on November 15, 2014

470 jour nal of P E A C E R E S E A RC H volume 45 / number 4 / july 2008

(CPIA). A high CPIA rating is generally asso- significant effects. Unfortunately, severe
ciated with faster growth and, in the post- autocracy appears to be highly successful in
conflict context, its effect on growth is maintaining the post-conflict peace. Again
atypically strong (Collier & Hoeffler, 2004a). using our benchmark of 40% risk, if the
Since growth is risk-reducing, there may thus polity is highly autocratic, the risk is only
be some presumption that an improvement 24.6%, whereas if it is not highly autocratic,
in the CPIA would indirectly be risk-reduc- the risk more than doubles to 62%. Clearly,
ing. A potentially important issue is whether we do not wish to advocate severe autocracy.
this is offset or indeed accentuated by any However, it is important to recognize two
direct effects of the CPIA on risk. In a variant, uncomfortable implications. First, when the
we therefore added the CPIA to the core international community exerts effective
regression, which of course already controls influence to prevent autocracy, risks of
for the growth rate.6 The CPIA was com- further violence are likely to substantially
pletely insignificant and the sign of the effect increase, and so other measures will be
was peace-promoting. While there is always a needed to offset this effect. The second impli-
danger that particular policies will be growth- cation is that international pressure for
promoting but inadvertently exacerbate risk, democracy should be justified by criteria
the statistical evidence does not indicate that other than peace-strengthening. Democracy
this is usually the case. does not appear to be an instrument for
enhancing the durability of post-conflict
Political Variables peace.
We next turn to the political variables.7 We Our next political variable is post-conflict
first consider the degree of democracy. elections. Evidently, these are a very high-
We rely upon the standard classification of profile international policy, and so it is of par-
countries on a scale of autocracy and democ- ticular interest to investigate their effect on
racy, that of Polity IV. Because the scale is risk. We find that post-conflict elections have
ordinal, it is inappropriate to treat it as a con- statistically significant effects, but these are
tinuous variable. Rather, we search for signif- not straightforward. Elections shift the risk
icant break points. We find only one such between years rather than either raising or
break point, located within the range of lowering it. Specifically, an election reduces
autocracy. There is a significant difference risk in the year of the election, but increases it
between states in which the polity is highly in the year following the election. Presumably,
repressive, with an autocracy ranking of worse in the election year, antagonists divert their
than –5, and all other states.8 Apart from this, efforts from violence to political contest,
the degree of democracy or autocracy has no whereas once the election is concluded, the
losers have a stronger incentive to return to
6 The CPIA variable was made available to us by the LICUS violence. To illustrate the magnitude of the
Unit of the World Bank. effect, consider an election held in year 3. This
7 Sometimes there are missing data on these variables. reduces the risk in the year of the election
Rather than lose observations, we set missing observations
to 0 and introduce a dummy variable which takes the value from around 6.2% to around 3.4%. However,
of 1 if the data are missing. See Greene (2003: 60) for in the following year, the risk is increased from
details on this procedure, known as a modified zero-order
regression. The estimates of the coefficients on the
around 5.2% to around 10.6%. Evidently, the
dummies for missing values are of little interest (see Collier, net effect, taking the two years together, is
Hoeffler & Söderbom, 2004, for details), and we do not modestly to increase risk.
comment on them here.
8 Countries that Polity counts as ‘in transition’ (coded as In our core regression (column 1), the elec-
–66, –77 or –88) are coded as missing democracy data. tion variable imposes the same coefficients but

Downloaded from jpr.sagepub.com at UNIV TORONTO on November 15, 2014

 Paul Collier et al. P O S T-C O N F L I C T R I S K S 471

with opposite signs in the year of the election rather than as mechanisms for increasing the
and in the subsequent year. We arrive at this durability of the post-conflict peace. Third,
specification through various steps. First, we such elections generate a misleading signal of
allow the coefficients to be different for the calm. During the year of the elections, the
two years, and we distinguish between the society experiences a lull, but this is followed
first election and subsequent elections, since by a resurgence of risk. Post-conflict elections
there is speculation in the practitioner litera- are, thus, precisely the wrong ‘milestone’ on
ture that this distinction is important (Table I, which to base international exit strategies.
column 2). If anything, it is the adverse effect Our final political variable is post-conflict
of elections in the year following the election constitutional architecture. Most post-conflict
that is significant, rather than the favourable constitutions adopt a broadly unitary state,
effect in the year of the election. However, but a few constitutions grant substantial
both for the first election and subsequent elec- regional autonomy. Evidently, the granting of
tions, the pattern is similar: a favourable effect such autonomy is endogenous. However, we
followed by an adverse effect. In column 3, we might expect that the normal preference on
combine the election year and the subsequent the part of the politically powerful is for a
year by imposing a sign-reversal and investi- unitary state, so that regional autonomy is
gate whether the first election differs from granted only in situations that are too
subsequent elections. The hypothesis that demanding for hopes of a unitary state to be
these two coefficients are the same cannot be realistic. Thus, our priors are that the under-
rejected at conventional levels of significance. lying risk is likely to be greater in contexts in
In further variants, we have investigated which autonomy is granted. Because there are
whether any of the years prior to the election few cases of autonomy, the variable is not sta-
year also show significant effects. Since a post- tistically significant at conventional levels:
conflict election usually takes two years to in our core regression, it is significant only
organize, the event must be known to citizens at 15%. However, the effect is very large.
well in advance. However, prior to the year of Without autonomy, the decade-risk is 46.2%,
the election, there is no effect on risk. Possibly, whereas with autonomy, the risk falls to
there is so much uncertainty and fluidity in 12.2%. This may be a chance result, or it may
post-conflict situations that pre-announced be because situations where autonomy is
events with a long lead-time are heavily dis- granted are already fundamentally safer.
counted. We also investigated whether the However, the result is at least suggestive that
shifting effect depended upon the level of autonomy can be a helpful constitutional
income of the society or its ethnic composi- design feature in post-conflict situations.
tion. Neither was significant. We have investigated a range of other
There are three important policy implica- aspects of political design but have not found
tions of this result. First, elections should not any to be even marginally significant. In
be treated as a systemic solution to the acute Table I, we report one such variant, an index
problem of post-conflict risk.9 Second, as of economic freedom.
with democracy itself, post-conflict elections
should be promoted as intrinsically desirable Social Variables
Since issues of ethnicity loom large in post-
9 Caplan (2005) considers elections in post-conflict situa- conflict discourse, we investigated a range of
tions. He suggests that if local elections precede national variables that describe the ethnic composi-
parliamentary elections, national elections are less likely to
generate a focus for violence and thus a breakdown of the
tion of the society. None of these variables
peace. was significant. In Table I, we report the

Downloaded from jpr.sagepub.com at UNIV TORONTO on November 15, 2014

472 jour nal of P E A C E R E S E A RC H volume 45 / number 4 / july 2008

most familiar measure, ethnic diversity. Like the size of the society (its population or its
the other measures, this is insignificant, but, economy). We find that the effect is far more
to the extent that it has an effect, it is significant when expenditure is measured in
favourable: societies that are more diverse absolute terms. Underlying this choice is
have lower risks. some characterization of the deterrence of
The other variable we investigated was the rebellion. That it is the absolute size of mili-
size of the diaspora. We measured only the tary deployment that matters suggests that
diaspora in the USA and took this as a per- the size of the potential rebel force is not
centage of the population of the post-conflict strongly related to the size of the society, so
country. The rationale for confining the that deterrence has strong economies of scale.
measure to the USA is partly that of consis- Virtually all rebellions have to go through a
tent data, and partly that it avoids the phase of being small, even if they subse-
problem of having to aggregate immigrants quently grow to very different sizes. In this
across distinct cultural, economic and legal incipient phase, they can perhaps be deterred
environments. A large diaspora is sometimes by a peacekeeping force of given absolute size.
a consequence of a past history of conflict, In some post-conflict situations, peace-
but in this article, we have made no keeping forces are not deployed, and so
allowance for this endogeneity. At least deployment is potentially biased according
superficially, one might expect that larger to the level of risk. To an extent, we are able
diasporas proxy more severe conflicts, which, to control for this by including a dummy
in turn, might have higher risks of conflict variable that takes the value of 1 if troops are
reversion. In fact, we find that diasporas sig- not deployed. This variable is significant
nificantly reduce post-conflict risks, and this and negative: troops are posted to environ-
result seems less likely to be a spurious con- ments that are intrinsically more risky.
sequence of endogeneity. The result is some- Similarly, the scale of deployment may be
what surprising, since diasporas tend to be a systematically related to the level of risk. We
source of finance for politically more extreme have attempted to find good instruments for
organizations. The effect is quite large: dou- these decisions but have failed to do so.
bling the diaspora reduces the risk from 40% Therefore, we enter the expenditure on
to 32.8%. peacekeepers directly as an explanatory vari-
able, recognizing that there are remaining
Military Variables dangers of endogeneity.10 If endogeneity is a
Finally, we introduce military variables. problem, the likely direction of bias is pre-
Recall that military spending by the post-con- sumably that more peacekeepers may tend to
flict government has been found to be coun- be deployed where the risks are greater: this
terproductive (Collier & Hoeffler, 2006). For would clearly be the implication of the
the present study, the United Nations has dummy variable for non-deployment. Thus,
supplied us with comprehensive and detailed if peacekeepers were ineffective, there would
data on the deployment of international be a spurious correlation in which they
peacekeeping troops in post-conflict situa-
tions. From this, we have constructed a vari- 10 Sambanis (2008) examines the endogeneity of UN

interventions. His IV estimations indicate that the impact

able on the expenditure per year on of UN interventions is similar to the one obtained from the
peacekeeping troops, measured as the loga- uninstrumented regression models. He tentatively con-
rithm of dollars spent per year. Potentially, cludes ‘that bargaining inside the UN is too complex to
respond in a straightforward manner to a particular logic
peacekeeping expenditures could be mea- of intervention’; in other words, UN interventions can be
sured either in absolute terms or relative to treated as exogenous.

Downloaded from jpr.sagepub.com at UNIV TORONTO on November 15, 2014

 Paul Collier et al. P O S T-C O N F L I C T R I S K S 473

appeared to be increasing the risk of peace without military intervention would bring
collapse. An apparently unfavourable effect risks down quite substantially but leave them
of peacekeepers might therefore be spurious, dangerously high at 55.2%. While the ‘poli-
and an apparently favourable effect is likely tics supported’ package is at the high end of
to be biased downwards. both observed post-conflict growth and
We find that peacekeeping expenditures observed post-conflict peacekeeping, neither
significantly reduce the risks of further con- is an extrapolation beyond the range of
flict, the effect being significant at 2%. The observed experience. We should note that
effect is large: doubling expenditure reduces even with the ‘politics supported’ package,
the risk from 40% to 31%. risks of conflict reversion are high: unitary
states with non-autocratic regimes are inher-
Comparing Two Packages ently fragile in these settings. As a portfolio
As will be evident from our results, political decision, lending into such risks is daunting.
design does not appear to reign supreme as However, given the enormously high costs of
the mechanism for post-conflict peace. We conflict, the risk-reductions that economic
first simulate a policy package in which polit- reconstruction and military peacekeeping
ical design has primacy in the peace strategy provide are likely to be very cost-effective
and which might be thought of as ‘business (Collier & Hoeffler, 2004c).
as usual’. The political design is for a unitary
state that adopts sufficiently democratic
Conclusion and Implications
structures to avoid severe autocracy and holds
elections in the third post-conflict year. In this article, we have used all available his-
External peacekeeping expenditure is set at torical cases of post-conflict episodes to
the mean for post-conflict situations, but investigate the risks of conflict reversion by
economic issues are relegated to the back- means of hazard functions. Even with this
burner, so that the economy is stagnant. All comprehensive coverage, the past may not be
other variables are set at the same levels as for a very accurate guide to the future: a changed
our other simulations. The risk of reversion to international environment may be making
conflict within the first decade is predicted to peace more secure. Nevertheless, the past is
be astonishingly high, at 75.4%. We then all that we have to guide policy: unchained
investigate the risk of a package which might from experience, international action may
be thought of as ‘politics supported’. In this generate risks as well as reducing them.
package, the political variables are the same as Our approach has been exclusively statis-
in the ‘business as usual’ package, but exter- tical. This has evident limitations. Our
nal military assistance and economic recovery results are best interpreted in conjunction
are also given priority. Spending on external with in-depth case-study evidence, the two
military peacekeeping is set at four times the approaches being complementary. Neverthe-
average, and the economy is given priority, less, the statistical approach does add value,
achieving growth of 10% per year.11 The risk in part because the case-study approach
of conflict reversion falls dramatically to provides excessive licence both to interpret
36.7%. The decline in risk is achieved by the particular situations according to the priors
combination of policies. Economic growth of the particular researcher and to generalize
in an unwarranted manner from the partic-
11In our sample, the median growth rate is 2.3% and the ular. Indeed, since most conflict case studies
mean is 1.5%. Zero growth rate is at the 25th percentile in
the sample distribution, and 10% growth is at the 95th per- are done by political scientists, there is a
centile. natural tendency to overemphasize political

Downloaded from jpr.sagepub.com at UNIV TORONTO on November 15, 2014

474 jour nal of P E A C E R E S E A RC H volume 45 / number 4 / july 2008

choices as explanations. Lest we should be is not peace-enhancing and, indeed, leaves

accused of an equivalent bias towards eco- the typical post-conflict society severely
nomics, we should note that our biases are exposed to the risk of further conflict.
constrained by our method. An implication of our analysis is that the
Post-conflict situations are typically fragile. political solutions need to be supplemented
This conclusion does not need any advanced by robust economic and military external
statistical methods, as it is apparent from the assistance. Economic development does sub-
raw numbers. We have benchmarked on the stantially reduce risks, but it takes a long
average risk that a post-conflict society reverts time. International forces appear to be very
to conflict within the decade, namely 40%. effective in maintaining peace, but the actual
This is far higher than the risk faced by the scale of provision is low relative to what
typical low-income country, and so the inter- appears warranted, given the high costs of
national community is correct to focus explic- conflict reversion. Unpalatable as it may be,
itly on post-conflict situations as warranting peace appears to depend upon an external
distinctive engagement. military presence sustaining a gradual eco-
In this article, we have attempted to nomic recovery, with political design playing
deduce the rules of thumb which character- a somewhat subsidiary role.
ize current practice. These rules of thumb We have also found a simple and statisti-
essentially reflect an implicit practitioner cally strong relationship between the severity
model of the causes of conflict that stresses of post-conflict risks and the level of income
motivation: grievances must be addressed by at the end of the conflict. This provides a clear
a political arrangement in order to avoid con- and uncontroversial principle for resource
flict reversion. In contrast to this interpreta- allocation that might be of use to the new
tion of conflict as being explicable in terms Peacebuilding Commission of the United
of motive, recent theories of conflict have Nations: resources per capita should be
emphasized feasibility: rebellions happen approximately inversely proportional to the
where they are feasible. This theory would level of income in the post-conflict country.
explain the high rate of reversion to conflict To date, allocations have been massively
in post-conflict situations as reflecting its deviant from this simple principle.
atypical feasibility, and it would prioritize We should stress that any statistical analysis
economic and military instruments for secu- such as our own needs to be supplemented
rity as necessary supplements to the political by appropriate contextual knowledge, before
solutions sought by practitioners. being applied in any particular situation.
To the extent that our characterization is There are limits to how much the past can be
reasonable, our results give cause for concern. a guide to the future, and there are even
It is, of course, entirely commendable that stricter limits to how much a statistical analy-
the international community should encour- sis can extract from this past experience.
age inclusive political arrangements in post- Nevertheless, because post-conflict issues are
conflict situations: democracy and elections so burdened with ideology and political
are intrinsically desirable. However, the sys- glamour, an analysis based on statistics can
tematic effect of such political arrangements be a useful antidote to other potent influences.

Downloaded from jpr.sagepub.com at UNIV TORONTO on November 15, 2014

 Paul Collier et al. P O S T-C O N F L I C T R I S K S 475

Appendix A. Sample of Post-Conflict Periods

Duration of Censored
End of recorded peace period?
End of war / recorded peace period (0 ⫽yes,
Country Start of war Start of peace peace period (years) 1 ⫽no)

Algeria 28-Jul-62 15-Jan-63 07-Feb-92 29.1 1

Algeria 07-Feb-92 31-Dec-00 31-Dec-02 2.0 0
Angola 11-Nov-75 31-May-91 28-Oct-92 1.4 1
Angola 28-Oct-92 22-Nov-94 31-Dec-02 8.1 0
Azerbaijan 25-Dec-91 27-Jul-94 31-Dec-02 8.4 0
Bosnia and Herzegovina 01-May-91 21-Nov-95 31-Dec-02 7.1 0
Burundi 30-Apr-72 25-May-72 18-Aug-88 16.2 1
Burundi 18-Aug-88 22-Aug-88 23-Nov-91 3.3 1
Burundi 23-Nov-91 31-Dec-92 21-Oct-93 0.8 1
Burundi 21-Oct-93 31-Dec-98 01-Jan-00 1.0 1
Cambodia 29-Jan-93 01-Jul-97 31-Dec-02 5.5 0
Cameroon 01-Jan-60 31-Dec-61 31-Dec-02 41.0 0
Chad 01-Nov-66 16-Jun-71 22-Mar-80 8.8 1
Chad 22-Mar-80 07-Aug-88 01-Jan-98 9.4 1
Chad 01-Jan-98 31-Dec-01 31-Dec-02 1.0 0
Chile 11-Sep-73 15-Sep-73 31-Dec-02 29.3 0
China 15-Jan-67 01-Sep-68 31-Dec-02 34.4 0
Colombia 15-Sep-49 31-Dec-62 15-Mar-84 21.2 1
Colombia 15-Mar-84 31-Dec-93 01-Jan-98 4.0 1
Congo, Dem. Rep. 04-Jul-60 01-Sep-65 28-Jan-93 27.4 1
Congo, Dem. Rep. 28-Jan-93 04-Feb-93 08-Oct-96 3.7 1
Congo, Dem. Rep. 08-Oct-96 31-Dec-00 31-Dec-02 2.0 0
Congo, Rep. 05-Jun-97 31-Dec-99 31-Dec-02 3.0 0
Dominican Republic 25-Apr-65 01-Sep-65 31-Dec-02 37.4 0
El Salvador 01-Jul-79 01-Feb-92 31-Dec-02 10.9 0
Ethiopia 01-Jan-74 28-May-91 31-Dec-02 11.6 0
Georgia 25-Dec-91 14-May-94 31-Dec-02 8.6 0
Guatemala 01-Oct-66 12-Jul-72 12-Mar-78 5.7 1
Guatemala 12-Mar-78 13-Apr-84 31-Dec-02 18.7 0
Guinea-Bissau 01-Jan-98 31-Dec-98 31-Dec-02 4.0 0
India 01-Jan-85 31-Dec-93 01-Jan-99 5.0 1
Indonesia 15-Dec-56 31-Dec-60 31-Dec-02 42.0 0
Iran, Islamic Rep. 03-Sep-78 31-Dec-79 06-Jun-81 1.4 1
Iran, Islamic Rep. 06-Jun-81 03-May-82 31-Dec-02 20.7 0
Jordan 17-Sep-70 24-Sep-70 31-Dec-02 32.3 0
Lao PDR 19-Apr-63 15-Feb-73 31-Dec-02 29.9 0
Lebanon 13-Apr-75 13-Oct-90 31-Dec-02 12.2 0
Liberia 01-Dec-89 28-Nov-90 15-Oct-92 1.9 1
Liberia 15-Oct-92 19-Aug-95 05-Apr-96 0.6 1
Liberia 05-Apr-96 20-Aug-96 31-Dec-02 6.4 0
Mozambique 21-Oct-79 04-Oct-92 31-Dec-02 10.2 0
Nicaragua 01-Oct-78 18-Jul-79 18-Mar-82 2.7 1
Nicaragua 18-Mar-82 19-Apr-90 31-Dec-02 12.7 0
Nigeria 06-Jul-67 12-Jan-70 18-Dec-80 10.9 1
Nigeria 18-Dec-80 01-Jan-81 02-Feb-84 3.1 1

(Continued)

Downloaded from jpr.sagepub.com at UNIV TORONTO on November 15, 2014

476 jour nal of P E A C E R E S E A RC H volume 45 / number 4 / july 2008

Appendix A. (Continued)
Duration of Censored
End of recorded peace period?
End of war / recorded peace period (0 ⫽yes,
Country Start of war Start of peace peace period (years) 1 ⫽no)

Nigeria 02-Feb-84 04-Mar-84 31-Dec-02 18.8 0

Pakistan 25-Mar-71 02-Dec-71 23-Jan-73 1.1 1
Pakistan 23-Jan-73 31-Jul-77 04-Nov-94 17.3 1
Pakistan 04-Nov-94 31-Dec-95 31-Dec-02 7.0 0
Peru 04-Mar-82 31-Dec-95 31-Dec-02 7.0 0
Philippines 01-Jan-72 31-Dec-92 01-Jan-00 7.0 1
Philippines 01-Jan-00 31-Dec-01 31-Dec-02 1.0 0
Romania 21-Dec-89 22-Dec-89 31-Dec-02 13.0 0
Russian Federation 11-Dec-94 30-Apr-96 01-Jan-98 1.7 1
Russian Federation 01-Jan-98 31-Dec-01 31-Dec-02 1.0 0
Rwanda 15-Nov-63 06-Feb-64 30-Sep-90 26.7 1
Rwanda 30-Sep-90 04-Aug-93 06-Apr-94 0.7 1
Rwanda 06-Apr-94 18-Jul-94 01-Jan-98 3.5 1
Rwanda 01-Jan-98 31-Dec-98 31-Dec-02 4.0 0
Serbia and Montenegro 01-May-91 03-Jan-92 31-Dec-02 11.0 0
Sierra Leone 23-Mar-91 23-Apr-96 01-Jan-98 1.7 1
Sierra Leone 01-Jan-98 31-Dec-00 31-Dec-02 2.0 0
South Africa 01-Jan-89 31-Dec-93 31-Dec-02 9.0 0
Sri Lanka 06-Apr-71 16-May-71 25-Jul-83 12.2 1
Sri Lanka 25-Jul-83 31-Dec-86 25-Jul-87 0.6 1
Sri Lanka 25-Jul-87 24-May-90 31-Dec-02 12.6 0
Sudan 01-Oct-63 28-Feb-72 17-Nov-83 11.7 1
Sudan 17-Nov-83 31-Dec-92 01-Jan-95 2.0 1
Tajikistan 01-May-92 31-Dec-97 31-Dec-02 5.0 0
Thailand 01-Jan-70 31-Oct-73 31-Dec-02 29.2 0
Uganda 08-Oct-80 24-Apr-88 08-Feb-96 7.8 1
Uganda 08-Feb-96 31-Dec-01 31-Dec-02 1.0 0
Vietnam 01-Jan-60 06-Feb-65 31-Dec-02 37.9 0
Zimbabwe 28-Dec-72 28-Dec-79 31-Dec-02 23.0 0
Peace periods are censored at 31-Dec-02, or earlier if information on explanatory variables is missing.

Appendix B. Data Sources available to us by the LICUS Unit of the World

Bank. Until recently, CPIA scores were not
CPIA We measure policy by the World publicly available; only since 2005 are the
Bank’s Country Policy and Institutional scores listed in the World Development
Assessment indicator (CPIA). It ranges from Indicators.
1 (poor) to 5 (good). The CPIA measure of Democracy ‘Democracy’ is a dummy which
policy has 20 equally weighted components takes a value of 1 if the polity score takes a value
divided into four categories: (1) Macro- of greater than –5. Source for the polity data:
economic management and sustainability of http://www.cidcm.umd.edu/inscr/polity/.
reforms, (2) Structural policies, (3) Policies for Diaspora We used the data on the foreign-
social inclusion and (4) Public sector born population in the USA and divided
management. The CPIA variable was made these numbers by the total population in the

Downloaded from jpr.sagepub.com at UNIV TORONTO on November 15, 2014

 Paul Collier et al. P O S T-C O N F L I C T R I S K S 477

country of origin. Sources: US Census; US UN Expenditure UN via World Bank:

Yearbook of Immigration Statistics (2004). data available from the authors.
Economic Freedom Index ranging from 0
(low) to 10 (high). Source: http://www. References
freetheworld.com/.
Economic Growth Using World Bank Alesina, Alberto; Arnaud Devleeschauwer,
William Easterly, Sergio Kurlat & Romano
(2005) data for GDP per capita we calculated
Wacziarg, 2003. ‘Fractionalization’, Journal of
the annual growth rates.
Economic Growth 8: 155–194.
Elections Source: Banks’ Cross-National Ashdown, Paddy, 2007. Swords and Plowshares:
Time-Series Data Archive (http://www.scc. Building Peace in the 21st Century. London:
rutgers.edu/cnts/about.cfm). Weidenfeld and Nicolson.
Ethnic Fractionalization The ethnicity Becker, Gary, 1968. ‘Crime and Punishment: An
variable involves a weighted combination of Economic Approach’, Journal of Political
racial and linguistic characteristics. The frac- Economy 76: 169–217.
tionalization variable is computed as 1 minus Caplan, Richard, 2005. International Governance
the Herfindahl index of group shares, and it of War-Torn Territories: Rule and Reconstruction.
reflects the probability that two randomly Oxford: Oxford University Press.
selected individuals from a population belonged Collier, Paul & Anke Hoeffler, 2004a. ‘Aid, Policy
and Growth in Post-Conflict Societies’,
to different groups. Source: Alesina et al. (2003)
European Economic Review 48(5): 1125–1145.
http://www.nber.org/papers/ w9411. Collier, Paul & Anke Hoeffler, 2004b. ‘Greed and
GDP Per Capita We measure GDP per Grievance in Civil War’, Oxford Economic
capita annually. Data are measured in con- Papers 56(4): 663–595.
stant 1995 US dollars, and the data source is Collier, Paul & Anke Hoeffler, 2004c. ‘Conflicts’,
World Bank (2005). in Bjørn Lomborg, ed., Global Crises, Global
Primary Commodity Exports The ratio of Solutions. Cambridge: Cambridge University
primary commodity exports to GDP proxies Press (129–156).
the abundance of natural resources. The data Collier, Paul & Anke Hoeffler, 2006. ‘Military
on primary commodity exports and GDP were Expenditure in Post-Conflict Societies’, Econo-
obtained from the World Bank. Export and mics of Governance 7: 89–106.
Collier, Paul & Anke Hoeffler, 2007a. ‘Civil War’,
GDP data are measured in current US dollars.
in Keith Hartley & Todd Sandler, eds,
Regional Autonomy Autonomous regions
Handbook of Defence Economics, Vol. 2.
are coded as a 0/1 dummy. An autonomous Amsterdam: North Holland (711–739).
region is recorded if a source explicitly Collier, Paul & Anke Hoeffler, 2007b. ‘Unintended
mentions a region, area or district that is Consequences? Does Aid Fuel Arms Races?’, Oxford
autonomous or self-governing. Autono- Bulletin of Economics and Statistics 69(1): 1–27.
mous regions must be contiguous with the Collier, Paul; Anke Hoeffler & Dominic Rohner,
country to which they belonged, on the pre- 2007. Beyond Greed and Grievance: Feasibility
sumption that such regions would be more and Civil War, mimeo, CSAE, Oxford.
likely to impose a check on central govern- Collier, Paul; Anke Hoeffler & Måns Söderbom,
ment decisionmaking than would non- 2004. ‘On the Duration of Civil War’, Journal
contiguous regions. Furthermore, they must of Peace Research 41(3): 253–273.
Collier, Paul; Lance Elliot, Håvard Hegre, Anke
be constitutionally designated as ‘autono-
Hoeffler, Marta Reynal-Querol & Nicholas
mous’ or ‘independent’ or ‘special’. Federal Sambanis, 2003. Breaking the Conflict Trap:
Districts or Capital Districts do not count as Civil War and Development Policy, World Bank
autonomous regions. Disputed autonomy is Policy Research Report. Oxford: Oxford
not recorded. Source: Beck et al. (2001). University Press.

Downloaded from jpr.sagepub.com at UNIV TORONTO on November 15, 2014

478 jour nal of P E A C E R E S E A RC H volume 45 / number 4 / july 2008

Fearon, James & David Laitin, 2003. ‘Ethnicity, Walter, Barbara F., 1999. ‘Designing Transi-
Insurgency, and Civil War’, American Political tions from Civil War: Demobilization,
Science Review 97(1): 75–90. Democratization, and Commitments to
Gleditsch, Kristian S., 2004. ‘A Revised List of Peace’, International Security 24(1): 127–155.
Wars Between and Within Independent World Bank, 2005. World Development
States, 1816–2002’, International Interactions Indicators, data file.
30(3): 231–262.
Gleditsch, Nils Petter; Peter Wallensteen, Mikael
Eriksson, Marianne Sollenberg & Håvard PAUL COLLIER, b. 1949, DPhil in
Strand, 2002. ‘Armed Conflict 1946–2001: A Economics (University of Oxford, 1975);
New Dataset’, Journal of Peace Research 39(5): Professor of Economics (1993– ) and Director,
615–637. Centre for the Study of African Economies,
Greene, William H., 2003. Econometric Analysis. University of Oxford (1991– ). Former
Englewood Cliffs, NJ: Prentice-Hall. Director, Development Research Group,
Hartzell, Caroline & Matthew Hoddie, 2003. World Bank (1998–2003). Current main
‘Institutionalizing Peace: Power Sharing and interest: political economy of developing
Post-Civil War Conflict Management’, countries. Most recent book: The Bottom
American Journal of Political Science 47(2): Billion: Why the Poorest Countries Are Failing
318–332. and What Can Be Done About It (Oxford
Hirshleifer, Jack, 2001. The Dark Side of the Force: University Press, 2007).
Economic Foundations of Conflict Theory.
Cambridge, MA: Cambridge University Press.
Miguel, Edward; Shanker Satyanath & Ernest ANKE HOEFFLER, b. 1967, DPhil in
Sergenti, 2004. ‘Economic Shocks and Civil Economics (University of Oxford, 1998);
Conflict: An Instrumental Variables Approach’, Research Officer, Centre for the Study of
Journal of Political Economy 112(4): 725–754. African Economies, University of Oxford
Paris, Roland, 2004. At War’s End: Building Peace (1999– ). Former Senior Associate,
after Civil Conflicts. Cambridge: Cambridge International Peace Research Institute, Oslo
University Press. (PRIO). Main interest: political economy of
Sambanis, Nicholas, 2000. ‘Partition as a developing countries. Most recent publication
Solution to Ethnic War’, World Politics 52(4): (with Paul Collier): ‘Civil War’, chapter in
437–483. Handbook of Defense Economics (Elsevier).
Sambanis, Nicholas, 2008. ‘Short-Term and
Long-Term Effects of United Nations Peace
Operations’, World Bank Economic Review MÅNS SÖDERBOM, b. 1971, PhD in
22(1): 9–32. Economics (University of Goteborg, 2000);
Small, Melvin & J. David Singer, 1982. Resort to UNIDO Research Fellow, Oxford University
Arms: International and Civil War, 1816–1980. (2000–07); Associate Professor, Goteborg
Beverly Hills, CA: Sage. University (2007– ). Main interest: applied
Stedman, Stephen J.; Donald S. Rothchild & micro-econometrics. Most recent publication
Elizabeth M. Cousens, 2002. Ending Civil (with Arne Bigsten): ‘What Have We Learned
Wars: The Implementation of Peace Agreements. from a Decade of Manufacturing Enterprise
Boulder, CO: Lynne Rienner. Surveys in Africa?’ (World Bank Research
Observer, 2006).

Downloaded from jpr.sagepub.com at UNIV TORONTO on November 15, 2014