The causes of armed conflict is a most common focus among conflict researchers, and the Armed Conflict Dataset has repeatedly been used to analyze this problem. Previous versions have not contained an authoritative definition of what an onset is, and this has caused unnecessary confusion. With this release, we also provide users with a clear and user-friendly operational definition of conflict onset.
Note that these datasets are derived from version 3-2005 of the Uppsala/PRIO database and cover the 1946-2004 period.
The data are available in two versions, country-year and calendar-time. The country-year file is based on the Gleditsch/Ward definition of the international system, and will inform the user whether the year observed an onset or not. All users of the Uppsala/PRIO data will know that a key question in this respect is how many years on inactivity we must observe within one dyad before we code a new onset between previously active parties. The country-year file comes with three pre-defined cutoff points of two, five and eight years, which allows for limited robustness checks. Besides, the country-year dataset includes eight alternative operationalizations of onset of intrastate conflict.
The calendar-time data structure, in comparison, allows for much more thorough robustness checks. This structure, originally presented by Raknerud & Hegre (1997) in 'The Hazard of War: Reassessing the Evidence for the Democratic Peace', is based on observations of all independent countries on the day where one of them experienced an onset. Since the relevant control group is closely linked to the observation on an onset, several types of filters can be applied within the same dataset, in order to check for robustness along various dimensions.