Reporting of conflict fatalities in the UCDP Georeferenced Event Dataset: insights from Kosovo

Many studies rest on the assumption that Uppsala Conflict Data Program data allow for an unbiased approximation to the temporal and spatial variation of deadly violence across cases. This research note compares fatality data in the program's Georeferenced Event Dataset with the name-by-name compilation of victims in the Kosovo Memory Book for the period 1 January 1998-30 June 1999. The extent of underreporting of fatalities in the Georeferenced Event Dataset increases with conflict intensity. Events outside Kosovo and events with many fatalities are disproportionally covered. The Georeferenced Event Dataset is more likely to ignore an event when many other events occur on the same day. Surprisingly, international observer missions do not make an event more likely to be reported. Despite the mentioned problems, the dataset's high estimates of conflict deaths mirror the temporal and spatial variation of violence fairly well. Some of the inconsistencies and biases identified in the selected case plausibly also occur in other conflicts. Events that span several months or lack specific location data impede analysis and are more prominent in the dataset's entries for Kosovo than in the entire Georeferenced Event Dataset.