DANYLIW RESEARCH SEMINAR ON CONTEMPORARY UKRAINE

How we respond to a civil conflict depends on what we know about it. That, in turn, depends on where we get our information. Not every event is observable, and not every observed event is publicly reported. Information providers diverge in the events and actors that attract their attention. One source may focus disproportionately on violence by rebels, another may emphasize government operations, while a third may not attribute violence to any armed group at all. Selective reporting may happen for commercial or partisan reasons, or because the government controls the press and requires it. As a result, different sources offer different perspectives on a conflict, and how violence begins, perpetuates and stops. This variation constitutes reporting bias -- the systematic under- or over-reporting of events.

 

Using new event data from the ongoing armed conflict in Eastern Ukraine, we perform parallel analyses of media-generated event data from pro-government, pro-rebel and third party sources, to examine how reporting bias affects the empirical study of armed conflict. We investigate the extent to which different sources suggest different patterns of strategic interaction between warring sides, and advance different conclusions about the causes, location and timing of violence.

 

We find that actor-specific reporting bias can profoundly affect both statistical inference and public opinion. According to data from Ukrainian sources, rebels are more likely than the government to unilaterally escalate violence. According to rebel sources, the opposite is true. Both Ukrainian and rebel sources predict more violence in equilibrium than do Russian and international sources. Each perspective has its own implications for how different actors behave in war, the need for third-party intervention, and whether intervention should be neutral or one-sided.

 

To investigate the effects of actor-specific reporting biases on policy preferences, we conduct a survey experiment, drawing participants from two national populations (United States and India). We find that respondents in both countries tend to support intervention against whichever side is shown to be committing the violence. By contrast, event reports without information about specific actors and tactics reduce support for intervention, and increase support for a limited, impartial response. These findings suggest that, in addition to confounding the statistical analysis of conflict, reporting bias can mobilize support for and opposition to specific armed groups.