Explaining Unattributed Violence in Open-Source Terrorism Data across Attack-types

• In Event: Examining Terrorism Trends and Threats

Abstract

Open-source terrorism data is known for having large amounts of unattributed violence, which raises serious reliability concerns for perpetrator-level analysis. A major part of the problem lies in the reliance on media coverage and the selection process of media content production. However, there is a lack of empirical analysis to distil how exactly such media selection process operates in shaping the available open-source data. This paper fills in the gap by focusing on the event ‘newsworthiness’ aspect of media selection. Using data from the Global Terrorism Dataset spanning over 50 years (1970-2021), I conducted multilevel fixed-effect analyses to investigate how event ‘newsworthiness’ factors (such as fatality, victims' occupational/social status, international involvement) influence the probability of perpetrators' identities being "unknown" across four types of terrorist events: bombings, armed assaults, kidnappings, and assassinations. The results suggest that the 'newsworthiness' effect is substantial and attack-type specific. Higher fatality rates and attacks on military/police targets increase data availability for all types of attacks. Attacks against international/foreign targets, government targets, and the use of hot weapons also significantly affect data availability, but limited to certain types of attacks. Interpretations and implications are discussed.

Author

• Lu Liu, Peking University