A global risk assessment model for civil wars☆
Introduction
In recent years, much emphasis in academic research and international politics has been put on bringing civil wars to an end, and keeping and building the peace in volatile post-conflict situations. One of the main reasons for post-conflict intervention is the prevention of further violent conflict. At least partly due to an increase in international conflict management efforts to prevent civil wars, their number worldwide has substantially decreased from a peak in the early 1990s (Human Security Centre, 2005, Human Security Centre, 2006). To allocate scarce resources efficiently, target countries for preventive interventions need to be selected prudently. While potential supporters of a preventive intervention will also take their own interests into account (Bercovitch and Schneider, 2000, Gilligan and Stedman, 2003, Greig, 2005, Greig and Rost, 2005), the main criterion should be the level of risk a country faces for experiencing a civil war in the near future.
In this study, we assess the expected risk countries face to see a civil war erupt within the next 5 years. These risk assessments are generated from multivariate models that are based on theoretical and practical considerations about the factors that correlate with or precede civil war onset. Our aim is thus not further to examine the causes of civil war outbreak in addition to the empirical studies that already explore these causes (e.g., Fearon and Laitin, 2003, Collier and Hoeffler, 2004, Sambanis, 2004, Hegre and Sambanis, 2006). While this study is only a first step towards constructing a comprehensive early warning scheme, practical international conflict management would greatly benefit from such a model, providing policy makers with methodological guidance on how to allocate scarce resources.
After discussing the relevant literature on risk assessment, early warning and civil war, and laying out theoretical expectations about factors that likely are linked to or precede civil war onset, we construct logit and, for comparison, neural network models that take these factors into account. To generate in-sample predictions, we calculate the predicted probability of civil war onset over the next 5 years for each country-year (our unit of analysis). In addition to analyzing the quality of our models with ROC-curves, the highest predicted probabilities are compared to the actual occurrence of war onset. Next, we simulate a situation of genuine risk assessments by restricting the model and data to generate out-of-sample assessments for three test periods (2003–2007, 1998–2002, and 1993–1997). Again, we compare the predicted probabilities to the historical record of civil wars that broke out during each of these 5-year periods and analyze the predictive quality of the models. Finally, we capture the situation as of 2007 (the last year with fully available data) to generate risk assessments for 2008–2012. The current political and security situation in the highest-risk countries is briefly described, as well as observers’ perceptions of the risk that these countries slide into violent conflict. In the conclusion, we discuss strengths and limitations of the presented approach with a view to further improve and extend it in the future.
Section snippets
Early warning and risk assessment, civil war and human rights violations
Early warning and risk assessment models have been constructed and tested for a number of humanitarian catastrophes, using a variety of methods and data types. One can broadly distinguish between models based on qualitative and quantitative data (or those that combine the two), and, among the quantitative models, between those using events-data and those using standards-based data. Along with the methods used, different objectives have been pursued: Whereas qualitative and events-data models,
Building a global risk assessment model for civil war
In this section, we describe a series of steps towards generating global risk assessments for civil war onset, producing forecasts for civil war risk up to the year 2012. Our goal is to explore the possibility of applying social science research to inform practical conflict management. While we do not aim to test causal theory and although there are important differences between theory testing and forecasting, out-of-sample predictions, as several authors have argued, can also be used as a test
In-sample risk assessments
In a first step, based on Model 1a in Table 1, we produce in-sample risk assessments, compare them to the historical record, and analyze their predictive quality. Some of the results of Model 1a are in line with existing research on the onset of civil war, but there are also some surprisingly diverging findings, as well as some new ones. The risk of civil war onset decreases with a higher level of economic development. This is one of the most robust findings in the civil war literature (
Conclusion
There are today a number of early warning and risk assessment models, to warn of various potential humanitarian catastrophes, including drought, tsunamis and political instability. Yet, Kofi Annan, a former UN Secretary General, in a report on conflict prevention, found that the UN had made “no significant progress” in strengthening capacities for early warning, information collection and analysis and that it lacked the “capability to analyze and integrate data from different parts of the
References (59)
An introduction to ROC analysis
Pattern Recognition Letters
(2006)- Annan, K., 2006. Progress Report on the Prevention of Armed Conflict. Available from:...
- et al.
Improving quantitative studies of international conflict: a conjecture
American Political Science Review
(2000) - et al.
Who mediates? The political economy of international conflict management
Journal of Peace Research
(2000) - et al.
KNIME: the Konstanz information miner
- et al.
Wars and state capacity
Journal of the European Economic Association
(2008) - et al.
Mapping mass political conflict and civil society: issues and prospects for the automated development of event data
Journal of Conflict Resolution
(1997) The War Trap
(1981)- et al.
Forecasting Political Events – The Future of Hong Kong
(1985) - Carey, S.C., 2005. Patterns of authority and the escalation of intrastate conflict. Paper Presented at the ECPR General...
Forecasting in International Relations – Theory, Methods, Problems, Prospects
Greed and Grievance in Civil War
On the duration of civil war
Journal of Peace Research
Ethnicity, insurgency, and civil war
American Political Science Review
Dangerous states and endangered people: implications of life integrity violations analysis
Where do peacekeepers go?
International Studies Review
Armed conflict 1946–2001: a new dataset
Journal of Peace Research
No Other Way Out – States and Revolutionary Movements, 1945–1991
Early warning of disastrous population movement
International Migration Review
Early warning: conceptual and practical issues
Stepping into the fray: when do mediators mediate
American Journal of Political Science
Ethnopolitical rebellion: a cross-sectional analysis of the 1980s with risk assessments for the 1990s
American Journal of Political Science
Cited by (28)
Evaluating forecasts of political conflict dynamics
2014, International Journal of ForecastingReading between the lines: Prediction of political violence using newspaper text
2018, American Political Science ReviewChallenging the status quo: Predicting violence with sparse decision-making data
2022, International InteractionsForecasting conflict in Africa with automated machine learning systems
2022, International InteractionsInternational Law in the Transition to Peace: Protecting Civilians under jus post bellum
2021, International Law in the Transition to Peace: Protecting Civilians Under Jus Post Bellum
- ☆
The views expressed in this paper are those of the authors and do not necessarily reflect those of the United Nations or OCHA. Earlier versions of this study were presented at a convention of the International Relations section of the German Political Science Association, Darmstadt, Germany, 13–14 July 2007; the Annual Conference of the German Peace Psychology Association, Konstanz, Germany, 15–17 June 2007; and the Polarization & Conflict meeting, Gaillac, France, 7–9 June 2007. We are grateful for the suggestions we have received at these conferences as well as the comments from two anonymous reviewers. We thank Thorsten Meinl for his invaluable help with calculating the neural network models in this study.