First Things First: Assessing Data Quality before Model Quality

In: The journal of conflict resolution: journal of the Peace Science Society (International), Band 57, Heft 6, S. 1090-1108

Abstract

We address weaknesses in the Peace Research Insitute Oslo (PRIO) Battle Deaths Dataset, and as a result draw contradicting conclusions to those presented by Lacina and Gleditsch. Our analysis focuses on the availability of data on battle deaths within specific conflict-years and problems encountered when data from multiple types of sources are combined. We repeat Lacina, Gleditsch, and Russett's analysis of battle deaths over time, with an attempt to provide a more robust model and incorporate an estimate of the uncertainty present in the PRIO Battle Deaths Dataset. This reanalysis reveals that the data used to establish the PRIO Battle Deaths Dataset does not offer a clear answer as to whether battle deaths have decreased or increased since the end of the Second World War. We contend that while the PRIO Battle Deaths Dataset offers the most comprehensive assembly of battle deaths data available to date, it is not suitable for analysis across countries or over time.