Predicting armed conflict using protest data
| Published date | 01 January 2025 |
| DOI | http://doi.org/10.1177/00223433231186452 |
| Author | Espen Geelmuyden Rød,Håvard Hegre,Maxine Leis |
| Date | 01 January 2025 |
https://doi.org/10.1177/00223433231186452
Journal of Peace Research
2025, Vol. 62(1) 3 –20
© The Author(s) 2023
Article reuse guidelines:
sagepub.com/journals-permissions
DOI: 10.1177/00223433231186452
journals.sagepub.com/home/jpr
1225162JPR0010.1177/00223433231186452Journal of Peace ResearchRød et al.
research-article2023
Regular Article
Predicting armed conflict using
protest data
Espen Geelmuyden Rød
Department of Peace and Conflict Research, Uppsala University
Håvard Hegre
Department of Peace and Conflict Research, Uppsala University and Peace Research Institute Oslo
Maxine Leis
Department of Peace and Conflict Research, Uppsala University
Abstract
Protest is a low-intensity form of political conflict that can precipitate intrastate armed conflict. Data on protests
should therefore be informative in systems that provide early warnings of armed conflict. However, since most
protests do not escalate to armed conflict, we first need theory to inform our prediction models. We identify three
theoretical explanations relating to protest-repression dynamics, political institutions and economic development as
the basis for our models. Based on theory, we operationalize nine models and leverage the political Violence Early
Warning System (ViEWS) to generate subnational forecasts for intrastate armed conflict in Africa. Results show that
protest data substantially improves conflict incidence and onset predictions compared to baseline models that
account for conflict history. Moreover, the results underline the centrality of theory for conflict forecasting: our
theoretically informed protest models outperform naive models that treat all protests equally.
Keywords
armed conflict, prediction, protest
Introduction
Protest is a low-intensity form of political conflict that
can precipitate intrastate armed conflict. Data on pro-
tests should therefore be informative in systems that
provide early warnings of armed conflict. Conflict
early-warning systems have become much more
advanced over the past years, responding to a rapidly
increasing interest from decisionmakers. Performant
forecasting systems are important: they can facilitate
early action to prevent violence, mitigate the conse-
quences of armed conflict and increase public awareness.
High-quality protest data with live updating schedules
are readily available (ACLED; Raleigh et al., 2010), but
no studies have persuasively shown they can be useful to
predict armed conflict. The main reason for this, we
argue, is that most protests do not lead to armed conflict,
so naively adding protest data to machine-learning
models does not necessarily yield good results (see Hegre
et al., 2019: for prediction of conflict incidence with
protest data). To succeed in our forecasting task, we need
theory to identify which protests have the potential to
lead to armed conflict violence.
1
We show that simply
entering protest data in a forecasting model without
careful modeling of dynamics and context does not yield
good predictive performance. We, therefore, build on
pre-existing theory on how protests are related to armed
Corresponding author:
Espen Geelmuyden Rød: espen.g.roed@pcr.uu.se
1
For a discussion of the importance of theory for conflict forecasting,
see Goldstone et al. (2010); Ward, Greenhill & Bakke (2010);
Cederman & Weidmann (2017); Chiba & Gleditsch (2017); Blair
& Sambanis (2020).
4 journal of P R 62(1)
conflict and identify three broad theoretical explanations
as the basis for our forecasting models. The first under-
lines how protest-repression dynamics can pave the way
for armed conflict. The second and third explanations
focus on the political institutions and socio-economic
conditions that make armed conflict more likely in the
wake of protest.
We operationalize the theoretical explanations and lever-
age the tools in ViEWS (the political Violence Early-
Warning System, Hegre et al., 2019) to evaluate whether
protest models improve armed conflict onset and incidence
predictions. Our empirical analysis consists of nine model
specifications that capture various aspects of the theoretical
arguments. We use random forest algorithms to generate
predictions of state-based armed conflict at the subnational
level for African countries. Models are trained on data from
1997–2016 and predict for 2017–2019.
The evidence shows that forecasting models with infor-
mation on protest activity do not unequivocally improve
armed conflict predictions compared to baseline models
accounting for conflict history. In fact, naive protest mod-
els, which treat all protests equally, do worse than the base-
line, especially for conflict onset. However, theoretically
informed models that unpack protest-repression dynamics
are better than both the baseline models and naive protest
models. There is also ample evidence that the institutional
and economic context matters for the relationship between
protest and armed conflict. In contrast to previous predic-
tion efforts with protest data, our approach improves both
armed conflict onset and incidence forecasts at the subna-
tional level. The results illustrate the importance of build-
ing armed conflict forecasting models on solid theoretical
foundations.
Overall, the article contributes to peace and conflict
research by marrying innovations in theory, data collec-
tion and methods for forecasting purposes. We also make
several additional contributions. First, efforts to predict
and forecast political violence have entered the peace
research mainstream (Hegre et al., 2013; Beger, Dorff
& Ward, 2016; Hegre et al., 2017; Witmer et al., 2017).
Most forecasting models, however, rely on slow-moving
‘structural’ factors, such as income or political institu-
tions, to generate forecasts. Consequently, the models
can distinguish countries at risk of violence from those
that are not, but they do not help identify where and
when violence will break out. Identifying the location
and timing of violence in high-risk countries is crucial for
conflict mitigation and prevention. Our approach shows
that predictive performance of both conflict onset and
incidence at the local level can be improved by
combining structural factors with event data (Chadefaux,
2014; Chiba & Gleditsch, 2017; Mueller & Rauh,
2018).
Second, although our primary objective is to
maximize predictive performance, our approach provides
insights into the usefulness of the theoretical arguments
we adapt for forecasting purposes. Evaluating a model’s
ability to improve prediction for unseen data can com-
plement the ‘p-value’ framework of hypothesis testing
(Ward, Greenhill & Bakke, 2010; Schrodt, 2014).
Without claiming that we test theoretical arguments in
any rigorous meaning of the word, our results show that
theoretical explanations help improve the performance of
a conflict early-warning system.
Literature review
Forecasting of armed conflict was high on the agenda in
peace research in the 1960s and 1970s (e.g. Choucri, 1974).
This agenda has seen a renaissance over the past ten years,
along with a general surge of forecasting and machine-
learning techniques in most scientific fields (see Hegre
et al., 2017: for a review). The most well-known armed
conflict forecasting models (Goldstone et al., 2010; Ward,
Greenhill & Bakke, 2010; Hegre et al., 2013; Bowlsby
et al., 2020) are set at the country-year level. They are
mainly based on static variables such as income and popu-
lation. A few models provide forecasts at finer geographic
resolutions (Witmer et al., 2017; Hegre et al., 2019, 2021)
and some at a more precise temporal scale (Ward et al.,
2013; Ward & Beger, 2017; Blair & Sambanis, 2020).
Although armed conflict is the prediction target that has
received the most attention, relevant studies also seek to
forecast coups or irregular leader changes (Bell, 2016a;
Ward & Beger, 2017), unrest (Chenoweth & Ulfelder,
2017), or regime change (Morgan, Beger & Glynn, 2019).
Many studies forecast the onset of armed conflict,
typically defined as the first year/month of violence in
a country above a given threshold after a given number of
years/months below the threshold. Other projects fore-
cast the incidence of armed conflict, whether violence is
above the threshold irrespective of violence levels in the
periods just before. Modeling the onset of new conflict
reveals new information but is also a more challenging
task. Incidence models, moreover, allow forecasting ter-
mination of ongoing conflicts.
Models forecasting protests have been forwarded
(Gurr & Lichbach, 1986; Cadena et al., 2015; Chenoweth
& Ulfelder, 2017), but not many models use protests to
forecast armed conflict. Some make use of event data to
forecast conflict (Ward et al., 2013; Chiba & Gleditsch,
2journal of PEACE RESEARCH XX(X)
Get this document and AI-powered insights with a free trial of vLex and Vincent AI
Get Started for FreeStart Your Free Trial of vLex and Vincent AI, Your Precision-Engineered Legal Assistant
-
Access comprehensive legal content with no limitations across vLex's unparalleled global legal database
-
Build stronger arguments with verified citations and CERT citator that tracks case history and precedential strength
-
Transform your legal research from hours to minutes with Vincent AI's intelligent search and analysis capabilities
-
Elevate your practice by focusing your expertise where it matters most while Vincent handles the heavy lifting
Start Your Free Trial of vLex and Vincent AI, Your Precision-Engineered Legal Assistant
-
Access comprehensive legal content with no limitations across vLex's unparalleled global legal database
-
Build stronger arguments with verified citations and CERT citator that tracks case history and precedential strength
-
Transform your legal research from hours to minutes with Vincent AI's intelligent search and analysis capabilities
-
Elevate your practice by focusing your expertise where it matters most while Vincent handles the heavy lifting
Start Your Free Trial of vLex and Vincent AI, Your Precision-Engineered Legal Assistant
-
Access comprehensive legal content with no limitations across vLex's unparalleled global legal database
-
Build stronger arguments with verified citations and CERT citator that tracks case history and precedential strength
-
Transform your legal research from hours to minutes with Vincent AI's intelligent search and analysis capabilities
-
Elevate your practice by focusing your expertise where it matters most while Vincent handles the heavy lifting
Start Your Free Trial of vLex and Vincent AI, Your Precision-Engineered Legal Assistant
-
Access comprehensive legal content with no limitations across vLex's unparalleled global legal database
-
Build stronger arguments with verified citations and CERT citator that tracks case history and precedential strength
-
Transform your legal research from hours to minutes with Vincent AI's intelligent search and analysis capabilities
-
Elevate your practice by focusing your expertise where it matters most while Vincent handles the heavy lifting
Start Your Free Trial of vLex and Vincent AI, Your Precision-Engineered Legal Assistant
-
Access comprehensive legal content with no limitations across vLex's unparalleled global legal database
-
Build stronger arguments with verified citations and CERT citator that tracks case history and precedential strength
-
Transform your legal research from hours to minutes with Vincent AI's intelligent search and analysis capabilities
-
Elevate your practice by focusing your expertise where it matters most while Vincent handles the heavy lifting