Each of the ViEWS outcomes is operationalized as its own set of dependent variables. Three of these, state-based armed conflict (AC), one-sided violence (OS) and non-state armed conflict (NS) are derived from the UCDP Georeferenced Event Dataset (GED) methodology, which is also used as a source of historical data (Croicu and Sundberg, 2017). Thus, data are available at a very fine-grained level - individual incidents of violence (battles) are recorded, at a day and location resolution. The data have been aggregated to the country and PRIO-GRID level as described in Croicu & Hegre (2018) (pdf). Data quality for the historical record has been peer reviewed, and evaluated as sufficiently good even for ViEWS most disaggregated unit of analysis (PRIOGrid-month) (Weidmann, 2014).
Since ViEWS requires near real-time data for risk assessment and forecasting, we have initiated and implemented, together with UCDP, a near real-time expansion of the UCDP-GED (UCDP-Candidate). This expansion, aiming to deliver data at the end of each month for the preceding month, relaxes some of the strict (yearly) definitions that UCDP uses in GED, while in the same time attempting to maintain as much compatibility as possible with the historical GED record. As such, this dataset includes data that have a high likelihood of inclusion in UCDP-GED, but do not pass certain definitional thresholds. These relaxed criteria can be either have to do with information inaccessibility in a real-time context or with the calendar-year nature of the UCDP definitions. Thus, events with unclear types of violence, actors, incompatibilities, fatality numbers, unclear sources or poor geography are included. An evaluation of UCDP-Candidate against the final versions of UCDP-GED will be produced as soon as version 18.1 of UCDP-GED is available.
Since both UCDP-GED and UCDP-Candidate contain records of more imprecise resolution in both the spatial and temporal domains, a multiple imputation routine has been developed to address these events. The routine uses high resolution events (with dates and locations known) in the same conflict to create an empirical distribution of the probability of conflict across the ViEWS units of analysis and samples a number of imputations from this distribution.
ViEWS operationalizes the dependent variable in different ways, depending on the needs of each component model. As such binary outcome variables, event counts and sums of fatality estimates are computed for each ViEWS outcome, for each unit of analysis (each PRIOGrid cell, country, actor) for each month. Additionally, spatial, temporal and spatio-temporal lags are computed for each of the outcome variables, up to a depth of 12 months and 2 spatial steps (neighbours of neighbours). Derived measures are also computed, including distances to the nearest event, time since previous event etc. from each observation in the ViEWS datasets.
For data validation purposes, ViEWS makes use of similar datasets as UCDP-GED, having established translation routines adapting these datasets to fit ViEWS outcomes, and computing the same measures as above. Currently, ACLED (Raleigh et. al., 2010) is used for this purpose, while other datasets (such as ICEWS and OEDA Phoenix) are investigated for future inclusion.
The fourth ViEWS outcome, Forced Displacement (FD) is currently under construction.
- Croicu, Mihai and Ralph Sundberg (2017), UCDP Georeferenced Event Dataset, UCDP, Uppsala University, available at http://ucdp.uu.se/downloads/ged/ged172.pdf
- Raleigh, Clionadh, Andrew Linke, Håvard Hegre and Joakim Karlsen. "Introducing ACLED: an armed conflict location and event dataset: special data feature." Journal of Peace Research 47(5): 651-660.
- Weidmann, Nils B. (2014), On the Accuracy of Media-based Conflict Event Data, Journal of Conflict Resolution 59(6): 1129 - 1149