Introduction
Recently, the Environmental Assessment Institute of Denmark organized
a forum of experts including some Nobel laureates in Economics. The aim
of the so called Copenhagen Consensus was to commission leading scholars
to assess the greatest challenges faced by humanity. The panel of experts
examined the estimates and prioritized the challenges offering the best
potential cost-benefit opportunities.1
One of the
challenges considered in the Consensus was the increasing incidence of
civil wars. Indeed, there is now a fair amount of literature on how civil
wars hinder economic and human development. This is also the view of multilateral
institutions such as the World Bank, which gives growing importance to
the analysis of causes and consequences of civil wars in developing countries.2
The experts of the Consensus unanimously agreed with the importance of civil wars as a major threat to development. Nevertheless, they actually omitted civil wars from their list of priorities, pleading insufficient information. What sort of information is there available on the nature and dynamics of civil wars and why is it insufficient?
The existing
data
One of the first and most influential quantitative works on conflicts
is the Correlates of War project (hereafter COW) described in the pioneering
book by David Singer and Melvin Small.3 COW was the
first long-term cross-country dataset on the incidence and intensity of
both inter- and intra-state conflicts. Constantly updated, its use in
quantitative analyses of war is still widespread, although nowadays other
datasets are available. Alternatives are the Civil War Termination data,
the datasets built respectively by Michael Doyle and Nicholas Sambanis
and by James Fearon and David Laitin, and the Uppsala/PRIO dataset from
the Department of Peace and Conflict Research at Uppsala University and
the International Peace Research Institute at Oslo. But to varying extents
all of the later datasets build on COW.
The development of these cross-country datasets has supported a recent boom in the empirics of conflict, expanding our understanding of civil war and supporting policy advice on how to prevent and overcome these conflicts. However, a disturbing question is whether this advice rests on a weak empirical base, as implied by the conclusion of the Copenhagen Consensus. The available econometric findings have generated stimulating but inconclusive debates, and it may well be that the quality of the data is at fault. Surprisingly enough, until very recently no one seems to have posed the question: how good is the data we rely on?
Quality
of standard cross-country datasets
Despite their common origins, there is considerable variety among cross-country
datasets, beginning with the range of definitions of the object of study.
Most of them include a measure of intensity (number of battle-related
killings) but some omit intensity measures entirely and limit themselves
to listing different conflicts and the time span in which they took place.
Among those that do include a conflict intensity number, very few provide
time series and when they do they give quite wide ranges. Most datasets
just give aggregate numbers without underlying time series. In all cases,
the numbers are often poorly documented, which makes it difficult to place
great confidence in them.
In a recent paper, coauthored with Jorge Restrepo and Michael Spagat of
the Royal Holloway College of the University of London,4
we discuss these issues and test the quality of the cross-country datasets.
Here I summarize our simple approach.
Over the
past two years, we have developed a general methodology for the in-depth
measurement of conflict activity in a single conflict. We have applied
this methodology to the Colombian civil war and the result of this effort
is a detailed, high frequency time-series dataset (hereafter RSV) that
covers more than 21,000 conflict related events over the period 1988-2003.
For every event we record the date and the place of occurrence (at the
level of the township); whether there was a clash between two or more
forces or a one sided, uncontested attack (in which case we distinguish
the type of attack and the group responsible for it); and the number of
killings and injuries.5 The data provides a detailed
long-term picture of the temporal and spatial dynamics of the conflict
as well as the evolution of the various conflict activities and their
impact in terms of casualties. In building the dataset, we have greatly
benefited from the efforts of the Colombian NGOs Centro de Investigación
y Educación Popular and the Comisión Intercongregacional
de Justicia y Paz, who publish Noche y Niebla, a quarterly periodical
that lists events of political violence gathered from a large network
of priests and collaborators as well as from over 25 national and regional
newspapers.6 We complement this source with press
reports and code it into a dataset after applying our methodological filter:
We focus merely on civil war dynamics rather than the broader concept
of “political violence” not necessarily connected with the conflict.
We evaluate
the quality of the cross-country datasets on civil war by comparing their
Colombia figures with those of RSV. The latter can be considered a “control
dataset” for a sample of the former. Obviously, it is a sample of
one, but in the short run it is the only feasible sample, given the high
cost of building datasets with the level of detail and the degree of care
that RSV applies to Colombia. At this stage we do not know the extent
to which our conclusions can be generalized, i.e., whether we have sampled
an outlier.
Our comparison suggests that the cross-country datasets have significant
quality problems. There are two main issues to point out. First, there
is a tendency to underestimate the intensity of the Colombian civil war
in terms of the casualties it produces. This finding is especially meaningful
because it is difficult to argue that RSV, which rules out all non-conflict-specific
violent manifestations (e.g. criminality figures), overestimates the yearly
number of killings.
We compare
the annual averages for killing rates in Colombia of RSV and 12 of the
most important cross-country datasets of civil war. These averages are
significantly below the RSV figure for all but four datasets. Of these
four, two overestimates are actually very close to the RSV figures. The
remaining two datasets provide ranks in which RSV lies, but the intervals
are particularly wide.
The exercise
with the annual averages is necessary since very few datasets provide
actual yearly data. However, when possible we also compare RSV with the
datasets that have time series. Almost all of these report wide ranges,
making this comparison sometimes ambiguous. In spite of this, the majority
of the estimates are unambiguously large underestimates of the annual
intensity of the Colombian civil war compared to RSV. The unique case
in which there is an overestimation in one year appears to be an error
in the respective cross-country dataset. In the cases when the figures
are compatible with RSV, the ranges suggested by the datasets are very
wide. For 2002, for instance, one dataset underestimates the death rate
by 500 in one case and over 2700 in another.
The second main finding is that no dataset captures the actual dynamics
of the civil war. In fact, the most prominent trend in the recent evolution
of the Colombian war, namely its significant upsurge after 1996, is unreported
in the cross-country information.
The future
of civil war research
Understanding the nature, the dynamics and the consequences of increasingly
common civil wars is necessary to provide urgent public policy advice.
Over the last 20 years, the effort of quantitative researchers and social
scientists has focused on the development and use of cross-country datasets.
This approach has been useful and insightful. But civil wars are still
a black box. Significant further progress in civil war research will require
improvements and extensions of existing datasets and the development of
new ones along the lines of the micro-dataset approach described in this
article.
NOTES
1. For more, see www.copenhagenconsensus.com.
2. See the 2003 World Bank Policy Report. Breaking the Conflict Trap:
Civil War and Development Policy. Online at http://econ.worldbank.org/prr/CivilWarPRR/text-26671/
3. Resort to Arms: International and Civil Wars, 1816-1980. Beverly Hills,
CA: SAGE, 1982.
4. The Severity of the Colombian Conflict: Cross-Country Datasets vs.
New Micro Data. Online at http://personal.rhul.ac.uk/pkte/126/Pages/research_on_colombia.htm
5. See a detailed description of the dataset in Restrepo, Spagat and Vargas'
The Dynamics of the Colombian Civil Conflict: A New Dataset. Online at
http://personal.rhul.ac.uk/pkte/126/Pages/research_on_colombia.htm
6. The latest issues of the periodical are online in Spanish at www.nocheyniebla.org.
Juan F. Vargas is completing his PhD in Economics at Royal Holloway University of London.