Systematic State Bias in Conflict Events Data

• In Event: Limitations and Advances in Conflict Data

Abstract

The majority of international relations events compendia rely on news sources, particularly articles from newspapers and wire services. However, those articles themselves are based on announcements from state officials (e.g., press releases, news conferences, speeches). Our claim is that the articles exhibit “state bias”: core aspects of the articles – who did what to whom – closely track how those aspects are portrayed in the announcements, rather than how they might be portrayed by, say, a non-state actor. In this paper, we leverage two newly constructed data sets on US-Russian relations – of state announcements, and of human-coded newspaper articles – to demonstrate, via sentiment analysis and calculation of semantic similarity, an extremely high overlap in the framing of both US and Russian actions; and we then extend this analysis to show that even for the more diverse set of news sources that make up compendia such as Phoenix, ICEWS, and TERRIER, similar overlaps exist. Our findings suggest that state bias in conflict events data is systematic and, given current pipeline workflow, impossible to avoid.

Authors

• David Sylvan, Graduate Institute of International and Development Studies

• Ashley Thornton, Graduate Institute of International and Development Studies

• Jean-Louis Arcand, Graduate Institute of International and Development Studies