The building blocks of ViEWS will be a set of theoretically informed model components designed to forecast individual or subsets of the outcome variables for each of the units of analysis. These models will focus on three phases in the conflict process: structural factors that change slowly and indicate where preconditions for conflicts suggest a high risk; triggers or isolated events that lead to abrupt changes in conflict intensity; and escalation – processes where dynamics of violent conflicts unfold and intensify violence. The components will be molded over existing studies that cover the units of analysis in ViEWS – countries (e.g., Eck and Hultman, 2007; Hegre et al., 2015), geographic locations (e.g., Raleigh and Hegre, 2009), and actors (e.g., Nilsson, 2008; Wood and Kathman, 2015).
These tasks require new data collection. The aim will be to update as many predictors as possible with a three-month lag. These indicators will mostly be coded by hand and will require considerable effort. To ensure that the coding is manageable, the project may start with a limited number of variables, gradually extend coverage, and phase out variables that fail to improve forecasts.
Structural factors are essential for estimating where conflicts are likely to occur, but also to set the trigger and escalation events data discussed below into context. Examples of factors that will be used in the project are roads, size of the local population (Raleigh and Hegre, 2009), extent of poverty in a region or a country (Cederman et al., 2011), presence of transnational ethnic kin (Cederman et al., 2013a), or political institutions (Hegre et al., 2001). Data on transnational ethnic kin and ethnic settlement patterns will also inform our models of where displaced populations move to (Moore and Shellman, 2007).
Triggers are isolated initial events that set violent processes in motion. We will in particular focus on triggers related to the dynamics of political institutions and election-related behavior. For instance, competitive elections have been found to act as a catalyst for ethnic civil war (Cederman et al., 2013b) or conflict recurrence (Brancati and Snyder, 2013; Flores and Nooruddin, 2012), especially when fraudulent (Daxecker, 2012). In semi-democratic regimes, election violence is more likely where the incumbent is uncertain of electoral victory (Hafner-Burton et al., 2014). ViEWS will track changes to the setup of formal institutions, the dynamics of electoral contests, and irregular leadership changes and coup attempts. We will maintain data on electoral cycles and term limits on executive offices. Both can be projected into the future and may highlight periods of particular future risk, conditional on the institutional setups of countries. We will also explore how triggering events are picked up in financial markets in ways that make such market indices useful as early-warning signals.
Escalation refers to how violent events propagate in time and space, and spill over to new violence forms and involve new actors. Escalation involves intensification, perpetuation, and recurrence of violence; spill-over to neighboring locations; involvement of related actors; and triggering of other conflict types including population displacement. Violence self-perpetuates at the country level (Walter, 2004) and at the local level (Buhaug and Rød, 2006; Raleigh and Hegre, 2009). There is variation across conflicts, however. Fragmented movements are less effective than unitary actors in settling disputes (Cunningham, 2013). Groups excluded from peace agreements are more likely to continue to fight (Nilsson, 2008). Other models show how conflicts spread across borders (Cederman et al., 2013a), with for instance refugee flows as an important contributing factor (Salehyan and Gleditsch, 2006).
Other model components indicate how some forms of violence cause other forms of violence to escalate. For instance, state-based conflicts often give rise to one-sided violence (Eck and Hultman, 2007) and to non-state conflict (Fjelde and Nilsson, 2012). Violence against civilians is the most significant predictor of forced migration (Moore and Shellman, 2004; Davenport et al., 2003), and population displacements spark instability in the region in the locations they move to (Salehyan and Gleditsch, 2006). International peacekeeping operations can reduce the risk of conflict recurrence (Doyle and Sambanis, 2006). By containing the intensity of conflict, they also indirectly increase the likelihood of conflict ending (Hegre et al., 2015). We will incorporate data on peacekeeping operations and other military interventions in conflicts as suggested by Hultman et al. (2013; 2014), and Beardsley (2011), to improve our forecasts.
The UCDP conflict events data will be an important source of predictors covering these mechanisms. Our models at the country level will use indicators such as the number of rebel actors, patterns of fragmentation and alliances, the prevalence of inter-rebel fighting, and refugee flows. At the sub-national level, we will focus on the presence of excluded groups and conflict activity in border areas. At the actor level we focus on actor attributes, including whether groups experience splits, join alliances, their military strength, and types of conflict termination (victories, settlements, ceasefires and low activity; Kreutz, 2010).