The first sub-objective is to build the data-collection routines and simulation software necessary for running a near-real time early-warning system according to the criteria stated above. To satisfy our transparency and coverage criteria, we will rely exclusively on publicly available information from pre-existing datasets, our own collection efforts, and news reports.
Units of analysis: Subnational; actors; countries. The output from the model will be assessments of the risk and likely severity of conflict at each of three units of analysis: sub-national geographical locations, countries, and actors. To make risk assessments for fine-resolution geographical locations ViEWS will rely on the PRIO-GRID (Tollefsen et al., 2012; http://www.prio.org/Data/PRIO-GRID/), a standardized spatial grid structure at a resolution of 0.5 x 0.5 decimal degrees. To provide risk assessments on the actor level, ViEWS will focus on relevant dyads of all actors identified by the UCDP as participants in the political violence events they record. ViEWS will also provide risk assessments at the country level building on Hegre et al. (2013), in part to be able to account for hitherto unknown actors. We will generate probabilistic forecasts for the risk of occurrence of political violence as well as probability distributions for the severity of these events in terms of number of people killed or displaced in a given period. Temporally, the system will explore the combination of various resolutions (days, months, years).
Collection of conflict variables, population displacement, and automatic risk factor retrieval. ViEWS will collect data and provide warnings for all three types of conflicts recorded by the UCDP (http://www.pcr.uu.se/research/ucdp/definitions/): armed conflicts involving states and rebel groups (AC), one-sided violence against civilians (OSV), and non-state conflict between rebels or communal groups (NS). UCDP-GED records all three categories of violence in an event-based format with high-resolution temporal and geographical references that can be matched to PRIO-GRID and UCDP actors (Sundberg and Melander, 2013). We will also collect data on inflows of forcibly displaced populations (FD) throughout Africa, and identify the UCDP conflict that is most likely the cause of the displacement for each inflow recorded. The project will also gradually build a collection of risk factors and early-warning signals that can be retrieved automatically from Factiva and other internet resources along the lines suggested in Chadefaux (2014). A few additional indicators will also be incorporated into the system as a probe to explore integration of other swiftly changing signals available for automatic reading – in particular, financial and other economic indicators, such as stock market fluctuations, food- and energy prices, and short-term growth performance.
Programming simulation system and website. ViEWS is a complex system and requires extensive software tools to be bound together. We will write a set of Python routines that access data, runs dynamic simulations, performs validation, summarizes results, and publishes output to a dedicated website. Many of these routines exist in simple forms written for the forecasts presented in Hegre et al. (2013), but will be rewritten in order to accommodate much greater needs for efficiency and flexibility in ViEWS. Forecasts will be reported in the form of maps, graphs, and numbers. The website will account for the methods and data used in the project, and account for the estimated uncertainty.