Predicting (de-)escalation of sub-national violence using gradient boosting: Does it work?

Figures & data

Figure 1. Share of cell-months with recorded BRDs that did not observe violence the month before or after. Source: UCDP GED 20.1.

Table 1. Predictors. Sub-national level variables were calculated using the PRIO-GRID v.3(alpha) framework.

Variable	Description	Source
ged_best_(sb|ns|os)_sum12	UCDP GED Candidate Dataset sum of state-based, non-state, and onesided violence (respectively) over a moving 12-month window	(Hegre et al. 2020)
ged_best_(sb|ns|os)_sum12grwtr6	6-month moving average growth of each of ged_best_(sb|ns|os)_sum12	(Hegre et al. 2020)
quad(1–4) and protest	Count of ICEWS events at the country-month level, separate variables for (verbal cooperation, material cooperation, verbal conflict, material conflict, protest)	(Boschee et al. 2018)
homicide	Global Burden of Disease 2017 country-year homicide rates	(Global Burden of disease Collaborative Network 2018)
v2x_clphy	Varieties of Democracy v.10 physical integrity index	(Coppedge et al. 2020)
v2x_libdem	Varieties of Democracy v.10 liberal democracy index	(Coppedge et al. 2020)
eld2m	REIGN 2-month half-life time until election day at country-month level	(Bell 2016)
esch	SHDI expected years of schooling at entrance age	(Smits and Permanyer 2019)
msch	SHDI mean years of schooling of adults aged 25+	(Smits and Permanyer 2019)
gnic	SHDI gross national income per capita	(Smits and Permanyer 2019)
lifexp	SHDI life expectancy at birth	(Smits and Permanyer 2019)
glcs_(landcovertype)	GLC-SHARE share of land covered with artificial surfaces, cropland, grassland, shrubs, mangroves, trees, freshwater, and snow, respectively	(Latham et al. 2014)
capdist	cShapes v.0.6 distance to capital	(Weidmann, Kuse, and Gleditsch 2010)
bdist3	cShapes v.0.6 distance to nearest internal border	(Weidmann, Kuse, and Gleditsch 2010)
riverdist	GSHHG v.2.3.7 distance to nearest major river or lake	(Wessel and Smith 1996)
coastdist	GSHHG v.2.3.7 distance to nearest coast	(Wessel and Smith 1996)
ruggedness	Terrain ruggedness	(Shaver, Carter, and Shawa 2019)
temp	CRU TS4.04 monthly gridded temperature	(Harris et al. 2020)
pre	CRU TS4.04 monthly gridded precipitation	(Harris et al. 2020)
spei(1,3,6,12,24)	Standardized Precipitation- and Evapotranspiration Index (1,3,6,12,24)-month intervals respectively using CRU TS4.04	(Harris et al. 2020)
waterrisk	WRI Aqueduct 3.0 Composite Water Risk Index (static)	(Hofste et al. 2019)
pop_gpw_sum	GPW 4.0 gridded yearly population (interpolated from 5-year data)	(CIESIN 2018)
eth_(promoted|demoted|excluded)	ETH EPR count of groups (promoted|demoted|excluded) at country-month	(Vogt et al. 2015)
epr_(promoted|demoted|excluded)	GeoEPR count of groups (promoted|demoted|excluded) at grid-month	(Vogt et al. 2015)
epr_regional	GeoEPR presence of regional groups (static)	(Vogt et al. 2015)

All variables were lagged 6 months.

Figure 2. Performance evaluation based on model rank per metric. Colors in online version.

Figure 3. Performance metrics for predictions between 2014 and 2019. CCC (${\rho }_c$) is bound between [–1,1], r-squared is bound between [0,1], R-squared is bound between [–∞,1], while MSE and TADDA are bound between [0, ∞]. Higher is better for CCC, r-squared and R-squared, while lower is better for MSE and TADDA. Colors in online version.

Figure 4. Zoom of the forecasts for Nigeria/Cameroon/Chad border (A + C) and Mozambique/Tanzania border (B + D). A + B shows the xgboost model's forecasts for October 2020 and the sum of battle-related deaths over the last year (since August 2020) in red. C + D shows the predictions (red, green, and blue line) plotted against the actual (black line) for the same cells as in A + B from January 2017 until December 2019. Colors in online version.

Figure 5. Accumulated local effects. Colors in online version.

Figure 6. The maps show the forecasts from the xgboost model for (A) October 2020 and (B) March 2021.

Hegre, Håvard, Mihai Croicu, Kristine Eck, and Stina Högbladh. 2020. “Introducing the UCDP Candidate Events Dataset.” Research and Politics 7(3): 1–8.

 Google Scholar

Boschee, Elizabeth, Jennifer Lautenschlager, Sean O'Brien, Steve Shellman, and James Starz. 2018. “ICEWS Weekly Event Data.” Harvard Dataverse. Accessed November 15, 2021. doi:10.7910/DVN/QI2T9A.

 Google Scholar

Global Burden of Disease Collaborative Network. 2018. Global Burden of Disease Study 2017 (GBD 2017) All-cause Mortality and Life Expectancy 1950–2017. Seattle: Institute for Health Metrics and Evaluation (IHME). Accessed January 14, 2020. http://ghdx.healthdata.org/record/ihme-data/gbd-2017-all-cause-mortality-and-life-expectancy-1950-2017

 Google Scholar

Coppedge, Michael, John Gerring, Carl Henrik Knutsen, Staffan I. Lindberg, and Jan Teorell, 2020. “V-Dem Country-Year Dataset V10.” University of Gothenburg: Varieties of Democracy (V-Dem) project.

 Google Scholar

Bell, Curtis, Besaw, Clayton., Frank, Matthew. 2021. “The Rulers, Elections, and Irregular Governance (REIGN) Dataset.” Broomfield, CO: One Earth Future. Accessed June 26, 2020. https://oefdatascience.github.io/REIGN.github.io/.

 Google Scholar

Smits, Jeroen, and Iñaki Permanyer. 2019. “The Subnational Human Development Database.” Scientific Data 6 (1): 1–5. doi:10.1038/sdata.2019.38.

 PubMedGoogle Scholar

Latham, John, Renato Cumani, Ilaria Rosati, and Mario Bloise. 2014. Global Land Cover Share (GLC-SHARE) Database Beta-Release Version 1.0-2014. Rome: FAO.

 Google Scholar

Weidmann, Nils B., Doreen Kuse, and Kristian Skrede Gleditsch. 2010. “The Geography of the International System: The CShapes Dataset.” International Interactions 36 (1): 86–106. doi:10.1080/03050620903554614.

 Web of Science ®Google Scholar

Wessel, Paal, and Walter H. F. Smith. 1996. “A Global, Self-Consistent, Hierarchical, High-Resolution Shoreline Database.” Journal of Geophysical Research: Solid Earth 101 (B4): 8741–8743. doi:10.1029/96JB00104.

 Web of Science ®Google Scholar

Shaver, Andrew, David B. Carter, and Tsering Wangyal Shawa. 2019. “Terrain Ruggedness and Land Cover: Improved Data for Most Research Designs.” Conflict Management and Peace Science 36 (2): 191–218. doi:10.1177/0738894216659843.

 Web of Science ®Google Scholar

Harris, Ian, Timothy J. Osborn, Phil Jones, and David Lister. 2020. “Version 4 of the CRU TS Monthly High-Resolution Gridded Multivariate Climate Dataset.” Scientific Data 7 (1): 1–18. doi:10.1038/s41597-020-0453-3.

 PubMed Web of Science ®Google Scholar

Hofste, Rutger Willem, Samantha Kuzma, Sara Walker, Edwin H. Sutanudjaja, Marc F.P. Bierkens, Marijn J.M. Kuijper, Marta Faneca Sanchez, Rens Van Beek, Yoshihide Wada, Sandra Galvis Rodrıǵuez, and Paul Reig. 2019. “Aqueduct 3.0: Updated Decision-Relevant Global Water Risk Indicators.” World Resources Institute. Accessed November 15, 2021. https://doi.org/10.46830/writn.18.00146

 Google Scholar

CIESIN. 2018. Gridded Population of the World, Version 4 (GPWv4): Population Density, Revision 11.” Palisades, NY: NASA Socioeconomic Data and Applications Center (SEDAC). Accessed November 10, 2021. doi:10.7927/H49C6VHW.

 Google Scholar

Vogt, Manuel, Nils-Christian Bormann, Seraina Rüegger, Lars-Erik Cederman, Philipp Hunziker, and Luc Girardin. 2015. “Integrating Data on Ethnicity, Geography, and Conflict: The Ethnic Power Relations Data Set Family.” Journal of Conflict Resolution 59 (7): 1327–1342. doi:10.1177/0022002715591215.

 Web of Science ®Google Scholar