Event data are a powerful tool to analyse as detailed as possible the violent behaviour and interactions of armed actors within and across wars. The inclusion of civilian fatalities into the most current created event datasets provides the opportunity to investigate systematically civilian victimizations in civil wars. However, these datasets are often criticized due to its heavy reliance on media data, using the argument that gathering information from newspapers causes various types of biases. Despite these common critics I argue that specific challenges show up when coding one-sided violence - the purposeful use of violence against civilians by armed actors - from media reports.

In general, definitions of one-sided violence are complex. On the one side they comprise theoretically demanding and empirically challenging concepts. On the other side a certain set of information is required from the consulted news source in order to code accurately this type of violence.

I will present results from my own coding work. Due to the substantial comparison of two different types of event datasets certain challenges have become manifested. First, while the extent of selection bias is difficult to investigate it seems that individual cases of civilian victimization are much less reported than events in which a higher number of civilians were killed. Second, the description of events from media reports often does not provide sufficient information in order to distinguish intentional civilian targeting from civilians killed as a “by-product” of military interactions between at least two armed actors. This context uncertainty, especially in combination with imprecise coding rules, results in biased estimation of civilian victims in civil wars.