Predicting the escalation of conflict: A global forecasting approach to conflict escalation using big data
Lead Research Organisation:
University College London
Department Name: Political Science
Abstract
The key objective of this research project is to forecast conflict escalation of intra-state armed conflicts, such as Syria, Libya, or Egypt, and deliver forecasting software packages that can be implemented by users of this project. The project considers two dimensions of escalation: temporal and spatial escalation of armed conflicts. In the context of this project, conflict escalation is an increase in the number conflictual events (e.g. episodes of fighting) over time and space. In focusing particularly on conflict escalation, this research project makes a new and unique contribution to the prediction of intra-state conflicts by developing a split-population forecasting framework that is motivated by a novel formal theoretical argument of conflict escalation. The split-population approach allows to forecast which countries are at risk of conflict, and in a second step assesses when and where conflicts escalate.
Conflict forecasting has made great advances in last years. Two reasons account for this progress: (a) Better data and (b) improved methods. While much progress has been made in forecasting conflict occurrence, less work has been done predicting the escalation of armed conflicts. Hence, this research project builds upon existing models that explain when and where conflicts occur but focuses on two research questions:
(1) What are the conditions under which conflicts occur and escalate over time (temporal- intensity escalation)?
(2) What are the condition under which conflicts occur and escalate over space (spatial-intensity escalation)?
The research proposes a game-theoretical argument that escalation is most likely where political concessions cannot be credibly granted within the existing political institutions. In line with existing work in the civil war context, credible concessions become difficult under weak institutions, deeply engrained discriminatory practices, and unpopularity of existing leaders. These governments are at risk to conflict escalation, whereas governments that are able to make concessions are almost immune to violent conflict. Thus, if strong opposition groups challenge governments that are at risk, conflict will escalate. Hence, the main argument of this research proposal is that conflicts with strong and successful challengers escalate if the government is unable to make credible commitments to these groups.
The proposed research project makes a new and unique contribution to the prediction of intra- state conflicts by developing a split-population forecasting framework that is motivated by the outlined theoretical argument of conflict escalation. The split-population approach allows to forecast which countries are at risk of conflict, and in a second equation when and where the 'at risk' conflicts escalate. Thus, the forecasting approach mirrors the theoretical argument. It first identifies which countries are likely to have commitment problems and in a second step when and where they will escalate when strong challengers form.
To forecast forecast both, the occurence (if) and escalation of conflict (when and where), the project will leverage the Global Database of Events, Language, and Tone (GDELT), a constantly updated open-source conflict database that provides spatio-temporal information of up to 100,000 conflictual and cooperative events per day. Hence, the proposed project explicitly deals with statistical forecasting that takes advantage of "big data" and increased computational power. The proposed research addresses the demand of governments, non-governmental organizations and corporations to monitor and forecast social and political risks. It has a clearly impact oriented strategy by delivering software packages, web-based prediction tools, and implementation strategies for government, non-governmental, and corporate users.
Conflict forecasting has made great advances in last years. Two reasons account for this progress: (a) Better data and (b) improved methods. While much progress has been made in forecasting conflict occurrence, less work has been done predicting the escalation of armed conflicts. Hence, this research project builds upon existing models that explain when and where conflicts occur but focuses on two research questions:
(1) What are the conditions under which conflicts occur and escalate over time (temporal- intensity escalation)?
(2) What are the condition under which conflicts occur and escalate over space (spatial-intensity escalation)?
The research proposes a game-theoretical argument that escalation is most likely where political concessions cannot be credibly granted within the existing political institutions. In line with existing work in the civil war context, credible concessions become difficult under weak institutions, deeply engrained discriminatory practices, and unpopularity of existing leaders. These governments are at risk to conflict escalation, whereas governments that are able to make concessions are almost immune to violent conflict. Thus, if strong opposition groups challenge governments that are at risk, conflict will escalate. Hence, the main argument of this research proposal is that conflicts with strong and successful challengers escalate if the government is unable to make credible commitments to these groups.
The proposed research project makes a new and unique contribution to the prediction of intra- state conflicts by developing a split-population forecasting framework that is motivated by the outlined theoretical argument of conflict escalation. The split-population approach allows to forecast which countries are at risk of conflict, and in a second equation when and where the 'at risk' conflicts escalate. Thus, the forecasting approach mirrors the theoretical argument. It first identifies which countries are likely to have commitment problems and in a second step when and where they will escalate when strong challengers form.
To forecast forecast both, the occurence (if) and escalation of conflict (when and where), the project will leverage the Global Database of Events, Language, and Tone (GDELT), a constantly updated open-source conflict database that provides spatio-temporal information of up to 100,000 conflictual and cooperative events per day. Hence, the proposed project explicitly deals with statistical forecasting that takes advantage of "big data" and increased computational power. The proposed research addresses the demand of governments, non-governmental organizations and corporations to monitor and forecast social and political risks. It has a clearly impact oriented strategy by delivering software packages, web-based prediction tools, and implementation strategies for government, non-governmental, and corporate users.
Planned Impact
The proposed research has clear and precise deliverables that are aimed to generate impact though prediction informed policy- and business decisions. The developed prediction tools implemented in a software package can be used by the stake holders of the project to:
-Monitor armed conflicts and their escalation
-Forecast temporal and spatial armed conflict escalation
-Inform policy and business decisions
-Inform policy interventions
-Improve the visualization and communication of political forecasts
Conflict monitoring allows governmental and non-governmental organizations to gain insights to escalation patterns and prepare ad-hoc responses. In addition, the actual forecasting tools provide facilities to generate forecasts that allow for longer (1-6 months) preparation of early action and resilience building measures. Risk assessments about future escalation patterns are targeted to inform policy and business decisions. The deliverables strategy of this project through software and the project's development of visual forecast communication of is clearly focused on impact generation.
(1) Who will benefit from this research?
-Policy makers involved with international conflict and post-conflict stability measures (e.g. MoD, FCO, DFID).
-International organizations that focus on the prevention of conflict and enforcement of peace (e.g. UN, NATO).
-Non-government organizations that have personal in conflict prone areas (e.g. Medecins Sans Frontieres).
-Non-government organizations that deal with the consequences of conflict escalation (e.g. Food and Agricultural Organization of the UN).
-Corporate actors that what to condition their business decisions on risk assessment (e.g. energy companies invested in conflict prone areas, insurance companies, or logistic companies).
-Corporate actors that deliver risk assessments to third parties (e.g. Economist Intelligence Unit or Oxford Analytica).
(2) How will they benefit from this research?
-Increasing their ability to monitor conflict occurrence and escalation.
-Strengthening their ability to forecast conflict occurrence and escalation.
-Improve their communication of forecasts and risk to organizational members and the public.
-Enable forecast-based decision-making.
-Building adequate resilience capacities based on early warnings.
-Enhance scientific knowledge within the stakeholders' organizations.
-Monitor armed conflicts and their escalation
-Forecast temporal and spatial armed conflict escalation
-Inform policy and business decisions
-Inform policy interventions
-Improve the visualization and communication of political forecasts
Conflict monitoring allows governmental and non-governmental organizations to gain insights to escalation patterns and prepare ad-hoc responses. In addition, the actual forecasting tools provide facilities to generate forecasts that allow for longer (1-6 months) preparation of early action and resilience building measures. Risk assessments about future escalation patterns are targeted to inform policy and business decisions. The deliverables strategy of this project through software and the project's development of visual forecast communication of is clearly focused on impact generation.
(1) Who will benefit from this research?
-Policy makers involved with international conflict and post-conflict stability measures (e.g. MoD, FCO, DFID).
-International organizations that focus on the prevention of conflict and enforcement of peace (e.g. UN, NATO).
-Non-government organizations that have personal in conflict prone areas (e.g. Medecins Sans Frontieres).
-Non-government organizations that deal with the consequences of conflict escalation (e.g. Food and Agricultural Organization of the UN).
-Corporate actors that what to condition their business decisions on risk assessment (e.g. energy companies invested in conflict prone areas, insurance companies, or logistic companies).
-Corporate actors that deliver risk assessments to third parties (e.g. Economist Intelligence Unit or Oxford Analytica).
(2) How will they benefit from this research?
-Increasing their ability to monitor conflict occurrence and escalation.
-Strengthening their ability to forecast conflict occurrence and escalation.
-Improve their communication of forecasts and risk to organizational members and the public.
-Enable forecast-based decision-making.
-Building adequate resilience capacities based on early warnings.
-Enhance scientific knowledge within the stakeholders' organizations.
People |
ORCID iD |
Nils Metternich (Principal Investigator) |
Publications
Leventoglu B
(2018)
Born Weak, Growing Strong: Anti-Government Protests as a Signal of Rebel Strength in the Context of Civil Wars
in American Journal of Political Science
Beiser-McGrath J
(2020)
Ethnic Coalitions and the Logic of Political Survival in Authoritarian Regimes
in Comparative Political Studies
Metternich N
(2020)
Strategic rebels: a spatial econometric approach to rebel fighting durations in civil wars
in International Interactions
Steinwand M
(2022)
Who Joins and Who Fights? Explaining Tacit Coalition Behavior among Civil War Actors
in International Studies Quarterly
Metternich N
(2015)
Firewall? or Wall on Fire? A Unified Framework of Conflict Contagion and the Role of Ethnic Exclusion
in Journal of Conflict Resolution
Hegre H
(2017)
Introduction Forecasting in peace research
in Journal of Peace Research
Chiba D
(2015)
Every Story Has a Beginning, Middle, and an End (But Not Always in That Order): Predicting Duration Dynamics in a Unified Framework
in Political Science Research and Methods
Beger Andreas
(2017)
Splitting It Up: The spduration Split-Population Duration Regression Package for Time-Varying Covariates
in R JOURNAL
Hollenbach F
(2018)
Multiple Imputation Using Gaussian Copulas
in Sociological Methods & Research
Steinwand M
(2020)
Who Joins and Who Fights? Explaining Tacit Coalition Behavior Among Civil War Actors
in SSRN Electronic Journal
Description | This project developed a forecasting framework to the prediction of rebel fighting escalation (RebelCast), which is based on data generated by RebelTrack. The project has provided key theoretical, methodological, and applied findings that have enabled the project to contribute to important academic debates and impact how government agencies conduct forecasts of political violence. Theoretical: The key theoretical insights of this project is that understanding the interdependency between armed actors is key in predicting political violence. We demonstrate that rebel organizations escalate violence in line with competition dynamics. We provide a new actor-centric approach to explicitly model strategic interdependence in multi-actor civil wars. However, escalation dynamics are not only dependent on other armed groups, but also the behavior other anti-government organizations, transnational conflict events, and previous conflict behavior. Hence, this project contributed to a growing literature that takes into account the spatio-temporal dynamics of conflict processes. The theoretical investigation also enabled new research collaborations with scholars from Uppsala University that will lead to a research stay during my upcoming sabbatical in 2019. Methodological: The project worked on a number of methodological challenges to big data prediction. First, the project helped to develop a framework for missing data imputation for time-series-cross-section data with a large number of observations. This is fundamentally important to any prediction approach, because we would leverage as much data as possible even for cases that don't have complete data. The second methodological key finding is how to evaluate classification predictions (binary model and count models) in the temporal context. We demonstrate that standard classification metrics for binary outcome data are prone to underestimate model performance in a binary-time-series-cross-section context. We argue for temporal residual based metrics to evaluate cross-validation efforts in binary-time-series- cross-section. Methodological I was able gain significant new skills relating to big data management (further enabled by a workshop of big data which was part of the grants training objectives) and computational methods. Application: The project has developed RebelTrack and RebelCast. RebelTrack and RebelCast focus on the prediction of rebel behavior. It provides a novel perspective to forecasting the spatial and temporal forecasting of rebel organization behavior by focusing on the strategic interdependencies between rebel organizations and government actors. RebelTrack is an R package leveraging existing event datasets, to provide roughly 140 measures on rebel organization behavior. In addition, the project delivers RebelCast a statistical learning framework to forecast rebel organization escalation. The innovation of RebelCast is that it the prediction of violence does not focus on a geographic unit (e.g. country or grid), but on the actors who are actually conducting armed conflict. This enables the prediction model to explicitly focus on the strategic dependence between actors and improve predictive accuracy. We implement an ensemble learning approach to the prediction of rebel escalation that accommodates the unique challenges of independent rebel organization behavior. The project finds that machine learning approaches that allow for the direct modeling of interdependencies and can differentiate between active and non-active phases of rebellion are providing improved predictions over current modeling approaches. Thus, purely geographic prediction approaches to conflict, which are most common in my discipline, will become more actor-centric based on my findings. |
Exploitation Route | The project enabled my to foster my relationships with the European Commission's conflict forecasting project Global Conflict Risk Index. The my project's findings have provided the foundation for suggestions to the Global Conflict Risk Index that have been integrated in their forecasting framework. This includes a) modeling high and low risk areas of conflict separately, b) moving to a more machine learning based approach of forecasting, c) moving towards subnational indicators of conflict, d) including measures of inequality in the forecasting model. This integration of key findings from my project will continue over the next years. I have also been invited to contribute to the German government's forecasting efforts starting in 2019. A major challenge working with UK government institutions on forecasting models, is the level of technological know-how. For example, the EU has dedicated engineers, political scientists, and statisticians to their implementation efforts of prediction models, while the UK government has no such resources and depends more heavily on external support. Hence, implementing methodically complex modeling approaches and getting the necessary political buy-in was much harder for me than on the EU level. |
Sectors | Agriculture, Food and Drink,Financial Services, and Management Consultancy,Government, Democracy and Justice,Security and Diplomacy |
Description | The project ``Predicting the escalation of conflict: A global forecasting approach to conflict escalation using big data'' has created non-academic impacts in different areas. I will differentiate between impact on society, national government sector, and international governmental sector. The project has made strong efforts to engage with the public. This has been achieved by several public events that spoke to members of the general public, but also government employees. This engagement has lead to greater awareness of the project as a whole, which has lead to an invitation to the Guardian's Science Podcast and talk about the prediction of conflict. The award-winning Science Weekly podcast is probably the most important science podcast in the United Kingdom and listeners around the world learn about big discoveries and debates in the sciences. I was interviewed by Ian Sample, who is one of the leading science journalists in the United Kingdom. I have also been consulted by the Economist for their data feature on the relationship of civil war and democracy. The main impact in regard to the government sector has come from my involvement with the European Union Commission's Global Risk Index. The project enabled me to support the EU with technical expertise an we have enabled and pushed the European Commission to implement machine learning techniques and focus on scientifically standardized measures of conflict. The Global Risk Index is the European Union's major quantitative risk assessment instrument to inform policy decisions. Another contribution we made to the Global Risk Index is that I suggested to include ethnic measures to forecast conflict. These were included and are now part of ongoing forecasts. There has been continued interest by the EU commission to update their forecasting framework and we are currently developing a machine learning package for this purpose. This has been presented at the EU commission in June 2018. The European Commission has now included machine learning approaches to conflict that were suggested to the European Commission in my presentation in June 2018. The next big push is to move the European Union to focus on more actor-centric approaches to forecasting. One of the major contributions of this project is that actor-centric forecasting can provide more accurate predictions of conflict. I have also been invited to the German Foreign Office to report on the findings of my research project and I am currently in discussion with the German Foreign Office on how to be collaborate in the future. At this occasion results were presented to the Netherlands Ministry of Affairs and members of the World Bank. I have also presented the findings of the research project to Ministry of Defense officials at the Alan Turing Institute. I am currently engaging in grant proposal (ESRC/Alan Turing) to further these relationships. The forecasting efforts of this project also allowed me to engage with the UNHCR who are interested in predicting, which refugees are most in need for targeted support. We are currently developing a prediction approach to identify households that are most vulnerable and will need support even in periods where budgets are being cut at the UN level. This work is likely to contribute to a new targeting scheme potentially affecting all refugees registered by the UNHCR. In the context of the Yemen Conflict, I have been in contact with UN agencies and Foreign Office officials about how to be provide prediction models on conflict escalation. These conversation happened in the context of a public policy forum at NYU, which I was invited to presenting findings of my ESRC funded research. |
First Year Of Impact | 2016 |
Sector | Government, Democracy and Justice,Security and Diplomacy |
Impact Types | Societal,Policy & public services |
Description | Advisory Board Uppsala Conflict Data Project |
Geographic Reach | Multiple continents/international |
Policy Influence Type | Membership of a guideline committee |
Description | ESCWA: United Nations Economic and Social Commission for Western Asia/ Risk Assessment Framework |
Geographic Reach | Asia |
Policy Influence Type | Contribution to a national consultation/review |
Impact | ESCWA is setting up a new Risk Assessment Framework for the Middle East. In this context, I was asked to report about my research findings on actor-centered forecasting approaches. This was reported in in the guiding document for the new risk assessment framework. |
Description | Presentation of actor centric forecasting framework to German Foreign Ministry, WFP, UNDP, ECOWAS, and other organizations |
Geographic Reach | Multiple continents/international |
Policy Influence Type | Influenced training of practitioners or researchers |
Description | Technical Advisor Role for EU Commission's Global Conflict Risk Index |
Geographic Reach | Europe |
Policy Influence Type | Participation in a guidance/advisory committee |
Impact | Improved conflict prediction that informs diplomatic efforts by the European Commission |
URL | https://drmkc.jrc.ec.europa.eu/initiatives-services/global-conflict-risk-index#documents/1059/list |
Description | Technical Advisor for EU Commission Global Risk Scan Index |
Geographic Reach | Europe |
Policy Influence Type | Membership of a guideline committee |
Impact | Implementation of prediction framework that is guiding policy decisions at the EU commission level. |
Description | Mobilizing for and against Democracy (MoDe) |
Amount | krĀ 11,847,000 (NOK) |
Funding ID | 302965 |
Organisation | Research Council of Norway |
Sector | Public |
Country | Norway |
Start | 08/2020 |
End | 07/2023 |
Title | RebelCast |
Description | Program to predict and forecast rebel organization behavior. |
Type Of Material | Computer model/algorithm |
Year Produced | 2018 |
Provided To Others? | No |
Impact | Not yet |
Title | RebelTrack |
Description | RebelTrack provides information on rebel organizations. |
Type Of Material | Data handling & control |
Year Produced | 2018 |
Provided To Others? | No |
Impact | not yet |
Title | Split-population statistical package |
Description | Statistical package to estimate and predict split-population models. |
Type Of Material | Data analysis technique |
Year Produced | 2015 |
Provided To Others? | Yes |
Impact | Models used as part of governmental conflict forecasting |
Description | Forecasting conflict networks |
Organisation | Uppsala University |
Country | Sweden |
Sector | Academic/University |
PI Contribution | We are investigating the potential of network based forecasting models that link strategic actor behavior to geographic unit forecasts. I provide technical expertise and theoretical knowledge on forecasting models based on network models. |
Collaborator Contribution | The partner provides database and expertise in geographic models of conflict forecasting. |
Impact | Planned research papers and integration in Views forecasting platform. |
Start Year | 2019 |
Description | Member of Methodology Working Group of the Global Conflict Risk Index of the European Commission |
Organisation | European Commission |
Country | European Union (EU) |
Sector | Public |
PI Contribution | Consulting new prediction and monitoring tool of the European Commission |
Collaborator Contribution | Development and implementation |
Impact | This project is implemented and generating predictions for policy makers. |
Start Year | 2014 |
Description | Project on Terrorism with the Santa Fe Institute |
Organisation | Santa Fe Institute |
Country | United States |
Sector | Academic/University |
PI Contribution | co-investigator on terrorism project |
Collaborator Contribution | co-investogator in project |
Impact | Currently working on draft paper |
Start Year | 2017 |
Description | Secretary of the Methodology Section of the Political Science Association |
Organisation | University of Nottingham |
Department | School of Medicine |
Country | United Kingdom |
Sector | Academic/University |
PI Contribution | Promoting methodology in political science and organizing conference at University College London |
Collaborator Contribution | Promoting methodology in political science and policy making. Organizing annual conferences for the section. |
Impact | -Two conferences -Briefing PSA on the activities in the section |
Start Year | 2016 |
Description | Secretary of the Methodology Section of the Political Science Association |
Organisation | University of Oxford |
Country | United Kingdom |
Sector | Academic/University |
PI Contribution | Promoting methodology in political science and organizing conference at University College London |
Collaborator Contribution | Promoting methodology in political science and policy making. Organizing annual conferences for the section. |
Impact | -Two conferences -Briefing PSA on the activities in the section |
Start Year | 2016 |
Description | Secretary of the Methodology Section of the Political Science Association |
Organisation | University of Warwick |
Country | United Kingdom |
Sector | Academic/University |
PI Contribution | Promoting methodology in political science and organizing conference at University College London |
Collaborator Contribution | Promoting methodology in political science and policy making. Organizing annual conferences for the section. |
Impact | -Two conferences -Briefing PSA on the activities in the section |
Start Year | 2016 |
Title | RebelTrack R package |
Description | The RebelTrack software allows researchers to create feature for rebel organization behavior. This is based upon data drawn from the UCDP-GED database. |
Type Of Technology | Software |
Year Produced | 2018 |
Open Source License? | Yes |
Impact | NA |
Description | 2nd International workshop on Modelling for Resilience Assessment |
Form Of Engagement Activity | A talk or presentation |
Part Of Official Scheme? | No |
Geographic Reach | International |
Primary Audience | Policymakers/politicians |
Results and Impact | Updated policy makers on predicting civil war negotiation onset from protest events |
Year(s) Of Engagement Activity | 2017 |
Description | Conflict and Change (Public Engagement) |
Form Of Engagement Activity | A talk or presentation |
Part Of Official Scheme? | No |
Geographic Reach | Regional |
Primary Audience | Public/other audiences |
Results and Impact | The conflict and change group at UCL organized a public debate on current policy issues. I presented work on the effect of international organization in conflict situations. |
Year(s) Of Engagement Activity | 2019 |
Description | Podcast for Guardian's ScienceWeekly Podcast discussing the prediction of conflict |
Form Of Engagement Activity | A broadcast e.g. TV/radio/film/podcast (other than news/press) |
Part Of Official Scheme? | No |
Geographic Reach | International |
Primary Audience | Public/other audiences |
Results and Impact | From the Guardian's Science Weekly podcast: Experts have been fighting about fighting throughout the ages. While theories have emerged to explain why we fight, there isn't a consensus in the research. In general, theories of war miss the mark for some. So why do we fight? And what can science tell us? To explore what experts are saying, host Ian Sample sits down with Mike Martin, author of Why We Fight; Nils Metternich, reader in international relations at University College London; and Mark Pagel, an evolutionary biologist with the University of Reading. |
Year(s) Of Engagement Activity | 2018 |
URL | https://www.theguardian.com/science/audio/2018/apr/20/the-science-behind-why-we-fight-science-weekly... |
Description | Prediction workshop |
Form Of Engagement Activity | Participation in an activity, workshop or similar |
Part Of Official Scheme? | No |
Geographic Reach | International |
Primary Audience | Policymakers/politicians |
Results and Impact | This was a workshop on predicting conflicts involving policy makers from NATO, EU, and non-governmental organizations. |
Year(s) Of Engagement Activity | 2016 |
Description | Warwick Q-Step Spring Camp |
Form Of Engagement Activity | Participation in an activity, workshop or similar |
Part Of Official Scheme? | No |
Geographic Reach | National |
Primary Audience | Undergraduate students |
Results and Impact | Presentation on forecasting civil wars that sparked an intense debate about the opportunities and limitations of prediction in the social sciences. |
Year(s) Of Engagement Activity | 2017 |
Description | Workshop on forecasting |
Form Of Engagement Activity | Participation in an activity, workshop or similar |
Part Of Official Scheme? | No |
Geographic Reach | National |
Primary Audience | Undergraduate students |
Results and Impact | Invitation to present results from my grant related research to undergraduates at the Q-step center at the University of Nottingham. |
Year(s) Of Engagement Activity | 2017 |
Description | Workshop on strategic rebel coalition formation |
Form Of Engagement Activity | Participation in an activity, workshop or similar |
Part Of Official Scheme? | No |
Geographic Reach | International |
Primary Audience | Postgraduate students |
Results and Impact | Enable dialogue between researchers and communication with post-graduate students |
Year(s) Of Engagement Activity | 2015 |