Replication Data for: I still haven't found what I'm looking for: Predicting conflict fatalities and security-related incidents with Google Trends and Wikipedia data

Conflict forecasting has seen two recent developments: a shift to predicting continuous variables and a debate about the value of structural and procedural variables. This paper contributes to these efforts and proposes the category of salience variables in the form of Google Trends and Wikipedia data. Internet searches can be precursors of conflict intensity as a result of e.g. an increase in protests, violent behavior, or public announcements. Data are readily and openly available, updated in real time, and provide global coverage which makes it ideal for near-real time forecasting. Prediction targets are the number of security-related incidents and battle-related, non-state, and civilian casualties. I demonstrate the value of \textit{salience} variables using various out-of-sample windows and performance metrics on the country- and province-month level. I find evidence that \textit{salience} variables have considerable predictive power, outperform other commonly used variables, and are thus a valuable addition to the conflict forecasting toolkit.

冲突预测领域近期出现两项重要进展：一是转向连续变量预测，二是围绕结构变量与程序变量价值的争论。本文为这些研究方向做出贡献，提出了以谷歌趋势（Google Trends）和维基百科（Wikipedia）数据为形式的显著性变量（salience variables）类别。互联网搜索可作为冲突强度的前兆——例如由抗议活动增加、暴力行为或公开声明等引发的冲突强度变化。此类数据易于获取且公开可用，实时更新并覆盖全球，因此非常适合近实时预测。预测目标包括安全相关事件的数量，以及战斗相关伤亡、非国家行为体伤亡和平民伤亡。我通过国家和省份月度层面的多种样本外窗口及性能指标，验证了显著性变量的价值。研究发现，显著性变量具有显著的预测能力，优于其他常用变量，因此是冲突预测工具包的重要补充。